DEMOGRAPHY OF SMALL MAMMAL POPULATIONS IN LONGLEAF PINE
UNDERGOING RESTORATION
Except where reference is made to the work of others, the work described in this dissertation is my own or was done in collaboration with my advisory committee. This thesis does not include proprietary or classified information.
______Nicholas Wingate Sharp
Certificate of Approval:
______F. Stephen Dobson James B. Grand Professor Associate Professor Biological Sciences Forestry and Wildlife Sciences
______Michael S. Mitchell, Chairman Stephen L. McFarland Assistant Professor Dean Forestry and Wildlife Sciences Graduate School
DEMOGRAPHY OF SMALL MAMMAL POPULATIONS IN LONGLEAF PINE
UNDERGOING RESTORATION
Nicholas Wingate Sharp
A Thesis
Submitted to
the Graduate Faculty of
Auburn University
in Partial Fulfillment of the
Requirements for the
Degree of
Master of Science
Auburn, Alabama May 13, 2005
DEMOGRAPHY OF SMALL MAMMAL POPULATIONS IN LONGLEAF PINE
UNDERGOING RESTORATION
Nicholas Wingate Sharp
Permission is granted to Auburn University to make copies of this thesis at its discretion, upon request of individuals or institutions and at their expense. The author reserves all publication rights.
______
Signature of Author
______
Date
Copy sent to:
Name Date
iii
THESIS ABSTRACT
DEMOGRAPHY OF SMALL MAMMAL POPULATIONS IN LONGLEAF PINE
UNDERGOING RESTORATION
Nicholas Wingate Sharp
Master of Science, May 13, 2005 (B.S., University of Alabama in Huntsville, 2001) (B.S., Birmingham-Southern College, 1993)
92 Typed Pages
Directed by Michael S. Mitchell
Successfully restoring the ecosystem function of a small mammal community, as part of a holistic ecosystem restoration effort, requires focus on population dynamics that operate over multiple spatial scales. I participated in a replicated, manipulative experiment investigating the use of fire and fire alternatives to restore a longleaf pine ecosystem, formerly one of the most expansive and biologically diverse ecosystems of
North America. I conducted a four-year mark-recapture study, comprising 119,700 trap nights, to assess demographic responses of small mammal populations to habitat alteration. Survival of golden mice (Ochrotomys nuttalli) was affected negatively by fire.
Survival and recruitment of cotton mice (Peromyscus gossypinus) were affected differently by restoration efforts. Survival varied over three spatial scales. Recruitment responded to both on-site habitat alteration and availability of immigrants from source habitat outside restoration areas. Attempting to restore historic assemblages of animal
iv
communities requires assessment of changes in population demographics as they respond to habitat alteration at a local scale, within the context of the surrounding landscape.
v
ACKNOWLEDGEMENTS
I thank the many people who provided academic, logistic, and moral support. I
thank Dr. Michael S. Mitchell for his guidance, advice, assistance, and friendship. Dr.
James B. Grand provided much needed assistance with data analysis, explained and explored the intricacies of capture-recapture modeling, and introduced me to program
MARK. Dr. Stephen F. Dobson’s population ecology class was a critical introduction to the theories and concepts of population ecology, which later formed the basis for my hypotheses and approach. In the first conversation I recall having with Dr. Dobson, he suggested the three types of variation I could explore – time, space, and treatment – that became the core of my research. Sarah Converse taught me the finer points of program
MARK and her consultation greatly improved my data analysis. Jim Nichols and Gary
White also provided much needed insights about subject matter on the periphery of my comprehension. Judy Christian, Marjorie Gentry, Beverly Edwards, and Patricia
Staudenmaier provided outstanding administrative support that allowed me to focus on my research. I thank all of my lab mates in the USGS Cooperative Fish and Wildlife
Research Unit for the fertile intellectual atmosphere. In particular I thank Travis Folk, who provided me with a steady supply of publications that always contributed to my understanding of longleaf pine and population ecology. I thank Rhett Johnson, Dale
Pancake, and the rest of the crew at the Solon Dixon Forestry Education Center for the many ways they contributed to my research. Finally, I thank my family. Only they truly know how much I relied on their support.
vi
Style manual or journals used: Ecological Applications, Journal of Mammalogy
Computer software used: Microsoft Office 2003, Program MARK 4.1, SAS 8.2,
SigmaPlot 9.0
vii
TABLE OF CONTENTS
DEMOGRAPHY OF SMALL MAMMAL POPULATIONS IN LONGLEAF PINE
UNDERGOING RESTORATION
INTRODUCTION………………………………………………………... Page 1
CHAPTER 1: Survival and recruitment of cotton mice …………………. Page 4
in longleaf pine undergoing restoration
CHAPTER 2: Restoring the small mammal community
of a longleaf pine ecosystem. ……………………………. Page 48
CONCLUSION…………………………………………………………... Page 82
viii INTRODUCTION
Humans have been contending with fire in the longleaf pine forests of southeastern North America since they arrived on the continent. The first European visitors to the region encountered vast expanses of sparsely distributed longleaf pine with an open understory of grasses and forbs (Platt 1999)a. In this region where almost daily summer thunderstorms cause one of the highest concentrations of lightning strikes in the world, fire was a natural and regular occurrence. Because of its commercial importance, mainly as a source for naval stores, longleaf pine was the first tree species of North
America to be studied in detail (Frost 1993). Studies of fire in longleaf pine followed
(e.g., Hayward 1939b, Wahlenberg et al. 1939c). Despite a long history of fire studies, surprisingly little is known about the effects of fire on the small mammal community of the longleaf pine ecosystem.
Boyer (1964) is the only researcher to document the small mammal community of longleaf pine stands in Alabama. Komarek (1939), Arata (1959), and Layne (1974) reported the effects of fire on the small mammal community in longleaf pine stands.
Additionally, Shadowen (1963), Hatchell (1964), Hofstetter 1973), and Layne (1974) demonstrated an increase in the numbers of cotton mice captured after fire in southern pine forests. None of these studies were of adequate sample sizes or experimental designs to infer mechanisms that caused the observed changes. Although there have been few studies of the effects of fire on the small mammals of longleaf pine, there have been many studies of the effects of habitat manipulation on the small mammal community in a
1 wide range of vegetation communities (Ahlgren 1966, Beck and Vogl 1972), Sullivan
and Sullivan 1982, Masters et al. 1998). These studies included the effects of fire,
herbicide, mechanical alteration, or some combination thereof. Unfortunately, most of
these studies still failed to link changes in habitat to changes in the vital rates of species
(but see McMurry et al. 1996). Such a link is required if the effects of treatments, or
more generally, habitat quality, are to be accurately assessed.
I studied the effects of fire and fire alternatives on the small mammal community
living in longleaf pine stands in which the natural process of fire has been interrupted.
The study took place at Auburn University’s Solon Dixon Forestry Education Center
(Dixon Center) in south-central Alabama. This research was part of the Fire and Fire
Surrogate Study (FFS), a joint effort between the US Department of the Interior and the
US Department of Agriculture to identify effective methods for restoring the United
States’ fire-maintained ecosystems. I had three objectives, 1) determine the desired future condition of the small mammal community so that I could set goals for restoration
and judge the effectiveness of restoration treatments, 2) determine if restoration
treatments affected small mammals, and 3) link demographic rates of cotton mice
(Peromyscus gossypinus) to habitat defined at a local and landscape scale.
Fieldwork had already begun when I joined the project in the summer of 2002.
The experimental design and the trapping protocol had already been established. The
decision to follow Pollock’s robust design was particularly wise, as it provided several
advantages in the data analysis. I identified the need to develop a picture of the small
mammal community living in a fire-maintained savanna and initiated trapping in
Conecuh National Forest (Conecuh NF). Realizing that the trapping schedule missed the
2
breeding season entirely, I began trapping in the winter in hopes of collecting information about reproduction and juvenile survival. The addition of trapping in Conecuh NF and in the winter greatly expanded my insights, and improved data analysis in ways I had not foreseen.
To meet my objectives I conducted a capture-recapture study of the small
mammals living at the Dixon Center and in Conecuh NF. In Chapter 1, I present a detailed study of the cotton mouse population living on experimental units at the Dixon
Center. I tracked changes in survival and recruitment and compared competing models to evaluate variation in these rates over time, space, and treatment. I incorporated measures of available source habitat for cotton mice to infer immigration from outside the study area.
In Chapter 2, I broadened my perspective to consider the small mammal community. I described differences between the small mammal communities living in
hardwood-pine woodlands at the Dixon Center and a longleaf pine savanna in Conecuh
NF. I evaluated the effects of restoration treatments on golden mice and calculated
survival rates for cotton mice living in Conecuh NF, for comparison with those of mice
living at the Dixon Center.
a Except for the following, literature citations appear in Chapters 1 and 2.
b Heyward, F. 1939. The relation of fire to stand composition of longleaf pine forests.
Ecology 20:287-304.
c Wahlenberg, W. G., S. W. Greene, and H. R. Reed. Effects of fire and cattle grazing on
longleaf pine lands, as studied at McNeill, Mississippi. United States Department
of Agriculture Technical Bulletin No. 683.
3
RH: Demographics of cotton mice in longleaf pine
SURVIVAL AND RECRUITMENT OF COTTON MICE IN LONGLEAF PINE
UNDERGOING RESTORATION
Abstract: Decades of fire suppression policy have led to an annual epidemic of catastrophic forest fires across the United States. The Fire and Fire Surrogate study
sought to determine the most ecologically and economically efficient methods to reduce
fuel levels and restore ecosystem integrity in the nation’s fire-maintained ecosystems,
through a national network of research locations. Each location employed four replicated
treatments representing options for forest restoration: 1) control - no action was taken; 2)
burn - prescribed fire was used to reduce fuel levels and maintain a frequent fire regime;
3) thin - trees were removed to reduce tree density; 4) thin/burn - thinning was followed
by prescribed fire. At the Gulf Coastal Plain site, where longleaf pine is the fire-
maintained ecosystem being restored, I conducted a capture-recapture study and used
Pollock’s robust design to follow changes in the survival and recruitment of cotton mice
living on experimental units. I compared competing models to evaluate variation in
survival and recruitment over time, space, and treatment. I incorporated measures of
available source habitat for cotton mice with reverse time modeling to infer immigration
from outside the study area. The top ranked survival model contained only variation over
time, but the closely ranked second and third models included variation over space and
treatment, respectively. The top four recruitment models all included effects for
availability of source habitat and treatment effects. I concluded treatments affected both
4
survival and recruitment. Burning alone appeared to create habitat sinks, while
treatments combining fire with thinning or herbicide appeared to improve habitat quality.
Source habitat outside the study area provided immigrants to experimental units. Results
indicated that treatments might have been more effective had they been implemented
over a larger area, mimicking the fires that once burned over great expanses of longleaf
pine savanna.
Key words: capture-recapture, demography, ecosystem restoration, fire, longleaf pine,
Peromyscus gossypinus, small mammals, survival, recruitment, robust design, reverse
time
Catastrophic wildfire, the unforeseen result of decades of fire suppression policy, has caused the loss of life and millions of dollars in property damage, destroyed vast expanses of the nation’s parks and forests, and costs millions more each year to fight.
The problem is now receiving attention as foresters and land managers look for ways to
efficiently and safely reduce dangerous fuel levels in the nation’s forests. These efforts
will inevitably affect the wildlife living within the forests. If fuel reduction efforts are to be applied in an ecologically sound way, we must understand how these applications affect wildlife.
Understanding how management actions affect animal populations requires monitoring the vital rates that drive population growth. Traditionally, the goal of wildlife management has been to achieve some desired population size, or density, and monitoring efforts have focused on that demographic. Van Horne (1983) and Pulliam
(1988) demonstrated density is not necessarily an accurate indicator of habitat quality,
5
therefore the vital rates that determine changes in population size must be considered.
Capture-recapture techniques provide robust estimates of these rates through a variety of models (e.g., Williams 2002). The ability to incorporate environmental variables as covariates in capture-recapture modeling provides an opportunity to link demographic
estimation with another growing and important discipline – landscape ecology.
My research is part of the national Fire/Fire Surrogate (FFS) study, a
collaborative effort between the US Department of the Interior and the US Department of
Agriculture (Weatherspoon and McIver 2000). The goal of the FFS study is to determine
the most ecologically and economically efficient methods to reduce fuel levels and
restore ecosystem integrity in the nation’s fire-maintained ecosystems. The key to
achieving this goal is the implementation of an experimental design common to a
nationwide network of research locations. The FFS study employed four treatments
representing options for forest restoration: 1) control - no action was taken; 2) burn -
prescribed fire was used to reduce fuel levels and maintain a frequent fire regime; 3) thin
- trees were removed to reduce tree density; 4) thin/burn - thinning was followed by
prescribed fire. All treatments are replicated three times. These treatments were
intended to result in a common desired future condition (DFC) - a forest stand structure
and fuel level such that 80 % of the trees would be able to survive a wildfire in all but
extremely unfavorable weather conditions.
I measured the effects of experimental treatments on small mammals at the Gulf
Coastal Plain site of the FFS study. The study took place at Auburn University’s Solon
Dixon Forestry Education Center (Dixon Center) in south-central Alabama, where
longleaf pine (Pinus palustris) savanna is the fire-maintained ecosystem being restored.
6
The Dixon Center includes 2,130 ha of managed pine forestland. Sites selected for
experimental units were longleaf pine stands in which the natural process of fire has been
interrupted; hardwoods have grown into the canopy alongside the dominate longleaf pine and at the time of selection had a dense shrub understory dominated by yaupon (Ilex vomitoria). The Study Plan for the Gulf Coastal Plain site further defines the DFC as a stand structure like that of the longleaf pine savannas that once expanded across southeastern North America (Frost 1993): an open canopy dominated by longleaf pine trees and an understory of grasses and forbs with some shrubs (Outcalt 2002). Herbicide is commonly used in pine plantations of the southeast to reduce understory vegetation that would compete with pine trees, and could help achieve the understory of grasses and forbs specified by the DFC. At the Gulf Coastal Plain site a fifth treatment was added to the four FFS treatments; herbicide/burn - herbicide is applied to the shrubby understory followed by prescribed fire.
Early explorers to North America described expansive savannas featuring an open grass/forb understory and sparsely distributed longleaf pines forming open canopies (Platt
1999). The longleaf pine ecosystem supports a high diversity of plant species, making it
one of the most species-rich communities outside of the tropics (Peet and Allard 1993).
It is also home to the threatened gopher tortoise (Gopherus polyphemus) and indigo snake
(Drymarchon corais), and the endangered red cockaded woodpecker (Picoides borealis).
The pyrogenic grasses and other fire-tolerant plants of the longleaf pine
ecosystem evolved under a process of frequent fires ignited by the lightning strikes of
spring storms. Landers et al. (1990) estimated that these fires had to occur every 2-4 years to maintain the ecosystem. These fires would burn at low intensity, sometimes over
7
great distances, through the continuous grass understory (Frost 1993; Peet and Allard
1993). When fire is suppressed longleaf pine savannas succeed to hardwood forests.
Fuel loads build in the woody understory, which can lead to high intensity, crown-
scorching wildfires (Peet and Allard 1993). Interruption of the natural fire regime, as
well as conversion of longleaf pine stands for agriculture and timber plantations, has caused the loss of this expansive, unique, and species-rich ecosystem. Today longleaf pine occupies < 3% of its original range (Platt 1999).
I chose the cotton mouse (Peromyscus gossypinus) as a study subject because they
were present in all 15 experimental units, were readily captured, and were present in large
enough numbers to confidently estimate demographic variables. I also chose cotton mice
because their reported affinity for bottomland hardwood forests (Wolfe and Linzey 1977)
allowed me to easily map potential source habitat (Pulliam 1988) for incorporation into
my demographic analysis. Cotton mice are omnivores and habitat generalists, but their
optimal habitat is purported to be bottomland hardwood forests, swamps, and mesic or
hydric hammocks (bottomland hardwoods; Wolfe and Linzey 1977). This assertion has
never been supported with demographic data other than abundance.
Four studies of fire in southern pinelands demonstrated an increase in the number
of cotton mice captured after fire (Shadowen 1963, Hatchell 1964, Hofstetter 1973,
Layne 1974). None were of adequate sample size or experimental design to infer
mechanisms that cause this response. Layne (1974) suggested that a rapid increase in the
density of cotton mice was due to a population overload of cotton mice spilling into the
burned habitat, or perhaps due to increased availability of pine and other seeds and
insects on the burned area.
8
I expected the study treatments to affect components of the habitat that provide
food and shelter for cotton mice. Prescribed fire was intended to mimic the natural
process of wildfire, removing woody understory vegetation and allowing the herb layer to flourish, creating an open understory. Fire effects vegetation in several ways that should benefit cotton mice. Soft mast and seed production increase after fire (Van Lear and
Harlow 2000), grasses and forbs expand coverage of the forest floor (Brockway and
Outcalt 2000), and resprouting vegetation is more nutritious and palatable (Stransky and
Harlow 1981). The goal of thinning was to restore stands to a density and composition resembling that of fire-maintained stands by removing deciduous trees and an over- abundance of longleaf pine. I expected this would open the forest canopy and more light would reach the understory, allowing the shrub or herb layer to flourish. Garlon (Dow
AgroSciences, Indianapolis, Indiana), which was used in this experiment, targets woody vegetation but is safe against grasses. Brockway and Outcalt (2000) suggested herbicide
application might be a way to expedite the restoration of longleaf pine habitat by encouraging the spread of grasses and forbs while permanently removing competition from oaks and other invading woody vegetation.
I expected treatment effects on local availability of resources to create source/sink
dynamics (Van Horne 1983, Pulliam 1988), altering animal dispersal patterns. A source
is a habitat in which reproduction exceeds mortality. It is self-maintaining. A source
produces a surplus of individuals that emigrate from the habitat. Since excess individuals
are exported, density in sources is usually constant over several generations. A sink is
any habitat in which reproduction is not sufficient to replace individuals lost to mortality.
A population can be maintained in a sink only by immigration of surplus individuals from
9
a source. Density in sinks may fluctuate widely and may even be greater than that of
sources. The matrix of sources and sinks across the landscape determines the probability
that any one area receives immigrants.
My goal was to create a complete picture of changes in cotton mouse population demographics over the course of the study so I could evaluate whether the treatments affected cotton mice. I perceived three ways in which the processes of survival and recruitment could vary: temporally, spatially, and by treatment. I used a mark-recapture study and Pollock’s robust design (Pollock 1982, Kendall et al. 1995) to estimate apparent survival (survival and emigration) and abundance. The robust design is the preferred model for estimation using mark-recapture data because it provides the most precise, theoretically substantiated, estimates of survival and capture probabilities available (Williams et al. 2002).
I estimated recruitment (births and immigration) with an asymmetric reverse time
approach under multistate robust design modeling. Analyzing capture-recapture data in reverse time provides inference about the recruitment process (Pollock et al. 1974). The reverse time equivalent of survival is the seniority parameter, the probability that an animal was in the population at time i-1 given that it was caught at time i (Pradel 1996).
Recruitment is equal to 1- seniority. Seniority can also be interpreted as indicating the relative contributions of survival and recruitment to the population growth rate (Nichols et al. 2000). I took a novel approach to infer the immigration of individuals from outside the study area by incorporating measures of available source habitat for cotton mice in recruitment models. I evaluated hypotheses by using program MARK (White and
10
Burnham 1999) to compare competing models that incorporated temporal, spatial, and
treatment variation in estimates of survival or recruitment.
I hypothesized that the burn, thin/burn, and herbicide/burn treatments would
improve habitat for cotton mice by increasing local availability of resources, with the
thin/burn treatment causing the best improvement. I hypothesized that the thin treatment
alone would not affect components of the habitat that are important to cotton mice,
therefore the habitat quality of thin units would be similar to that of control units. Since
cotton mice are omnivores and habitat generalists, I hypothesized they would be able to
locate adequate resources to survive and reproduce on thin and control units. I also
hypothesized that bottomland hardwoods in close proximity to experimental units would
serve as sources, providing immigrants to those units and obscuring the effects of
treatments on population size. Based on my hypotheses, I predicted that: 1) population growth, survival, and recruitment would be highest on thin/burn units, followed by burn and herbicide/burn units, 2) population growth, survival, and recruitment on thin units
would differ little from that of control units, the size of populations on thin and control
units would not change dramatically from the beginning to the end of the experiment, and
3) recruitment would be higher in populations close to large areas of bottomland
hardwoods and swamps.
METHODS
Study area
The Dixon Center is located in Covington and Escambia counties in the Gulf
Coastal Plain of south-central Alabama (31°9’ N, 086°42’ W). Summers in this region
are hot and humid (often 35° C and 95% humidity), while winters are mild. Average
11
annual precipitation is 14.22 cm (Outcalt 2002). Most precipitation occurs during late
winter through spring and during afternoon showers in the summer. The terrain is
relatively flat with rolling hills and ranges in elevation from 30 m to 100 m (Outcalt
2002).
Longleaf pine dominates much of the managed forest of the Dixon Center, but
loblolly, slash, shortleaf (P. echinata), and spruce pine (P. glabra) are also abundant
(Fischer and Holler 1990). Hardwoods can be found interspersed throughout the
overstory and midstory in mixed pine-hardwood stands and concentrated along
bottomland streams. Laurel oak (Q. laurifolia), white oak (Q. alba), and water oak (Q.
nigra) are most numerous, while American beech (Fagus grandifolia), sweetgum
(Liquidambar styraciflua), hickory (Carya spp.), maple (Acer spp.), ash (Fraxinus spp.), and elm (Ulmus spp.) also occur (Fischer and Holler 1990). The understory at the Dixon
Center is dominated by yaupon, along with lesser amounts of blueberries (Vaccinium spp.) and gallberry (I. glabra; Outcalt 2002). In places yaupon forms shrubby thickets that are quite dense.
Experimental design
The experiment was laid out in a randomized block design, with three replicates
per treatment (Figure 1). Experimental units were forest stands approximately 15.2 ha in
size. Three blocks were arranged so that experimental units shared similar soil types and generally shared the same location. Each of the four treatments and a control were randomly assigned to one of five experimental units within each block for a total of 15 experimental units. Minimum distance between experimental units was 156 m.
12
Treatment and trapping schedule
All thinning took place between February and April 2002 (Figure 2). Burn and thin/burn sites were burned for the first time in April and May 2002. Herbicide was applied in September 2002. Herbicide/burn sites were burned in the spring of 2003. A second burning of the burn and thin/burn sites was conducted in the spring of 2004. Due to weather conditions, one burn site was not burned a second time until July, after the first trapping period of the summer had taken place. Trapping followed Pollock’s (1982) robust design. I conducted ten trapping sessions (primary periods), each comprising seven consecutive nights of trapping (secondary periods; Figure 2). I trapped on each experimental unit twice during the summer for the four year duration of the study, with approximately one month between the two summer periods. I began winter trapping in
February 2003 to gather more demographic data during the breeding season.
Small mammal trapping
I sampled the small mammal community using standard capture-recapture methods. I installed a 0.81 ha trapping grid on each unit, locating most trapping grids in the center of each experimental unit. In some cases I opted to offset grids so as to avoid dirt roads. I recorded the location of trapping grids with a handheld GPS unit. Each grid had 100 trapping stations, located at the intersection of 10 rows and 10 columns placed at
10 m intervals. I placed a single Sherman live trap, baited with sunflower seeds, at each trap station. I checked traps each morning. During the summer, I closed traps during the heat of the day and re-opened and baited them in the evening. During the winter, I placed batting in each trap for nesting material. Due to time constraints imposed by short winter days, I re-baited traps and left them open during the day.
13
I gave each newly captured animal a uniquely numbered ear tag (monel ear tags
from Western Tag Co., Salt Lake City, Utah) and recorded the sex, age (sub-adult or
adult, determined by pelage sensu Pournelle 1952), reproductive condition (evidence of
lactation in females, distended scrotum in males), weight, hind foot length, tail length,
capture location, and identification number of each captured individual. I released
animals at the point of capture.
Model selection and inference
Model selection and inference followed the methodology of Burnham and
Anderson (2002). I ranked models using Akaike’s Information Criterion corrected for
small sample size (AICc; Akaike 1973). The model with the lowest AICc value is
considered the most parsimonious approximation of truth among the set of candidate
models, given the data. I used model deviance, the difference in AICc values between the
top ranked model and each competing model (ΔAICc), and normalized AICc values
(Akaike weights, wi) to compare the relative performance of competing models. Akaike
weights can be viewed as the weight of evidence in favor of a particular model. The ratio
of Akaike weights (wi/wj), or evidence ratio, indicates the likelihood of one model over
another (Burnham and Anderson 2002).
Survival models
I used the robust design model in program MARK (White and Burnham 1999) to
examine the temporal and spatial structure of survival and to evaluate the effects of the treatments on survival. I defined the nine intervals between primary periods were defined in monthly units (e.g., 45 days = 1.5 month) so that survival would be estimated as a monthly rate. Each of the 15 experimental units was treated as a separate group. The
14
robust design in program MARK estimates apparent survival, temporary emigration, initial capture probability, and recapture probability. I used a logit transformation (Cooch and White 2004) to incorporate characteristics of individuals (individual covariates), such as age, with estimation of survival rates. I used the Huggins estimator (Huggins 1989,
1991) to incorporate individual covariates with capture and recapture probabilities. The
Huggins estimator does not include population size in the estimating equation.
Population size estimates are derived from the estimated capture probabilities.
Typically, model selection begins with a goodness of fit test to ensure a highly parameterized, global model adequately fits the data. Selection proceeds with identification of models that explain the data more parsimoniously than the global model.
The global model should include all effects thought to be relevant to the parameter to be estimated (Burnham and Anderson 2002). Presently, no suitable goodness of fit test exists for the robust design. Data were generally too sparse to support highly parameterized models of survival including interaction terms, so my most highly parameterized survival model approximated the effects of time, space, treatments, and age with additive model terms.
To determine the temporal structure of survival that best described the data, I fit models that estimated survival rates seasonally and periodically. For seasonal structure
(season), I defined the intervals between primary periods as spring, summer, or fall- winter, according to the time of year between primary periods, and calculated one estimate of survival for each of the three seasons. For periodic survival (period) I estimated a unique survival rate for each of the nine intervals between primary periods. I also included a model in which survival rates did not vary over time.
15
To determine the spatial structure of survival that best described the data, I fit
models that estimated survival on four spatial scales. First, I assumed each experimental
unit hosts a distinct population. I presumed estimates of survival reflect local availability
of resources (including competition among individuals), and therefore the population on
each experimental unit should have a different survival rate. I modeled this by including
an effect term for each unit (unit). Second, I defined a population as a group of
interbreeding individuals. Using the greatest distance dispersed by a cotton mouse that
has been reported, 853m (Wolfe and Linzey 1977), I used ArcGIS 8.3 (ESRI 2004) to
place a 900 m buffer around each trapping grid, and dissolved these buffers into one
another at points where they intersected. From this 900 m spatial configuration, I
designated five neighborhoods (neighborhood; Figure 1). Alternatively, I considered a
more localized configuration in which unit 15 was distinct from units 12, 13, and 14 (and
all other groups were the same as before). This latter organization occurs at a maximum
buffer distance of 570 m and constitutes the third spatial configuration (locale). Finally, to model the population as continuous across the study area, I included models with no spatial variation in survival estimates.
I modeled treatments in three ways: 1) I handled each treatment as a separate
effect on survival (trt), 2) I constrained survival rates on thin units to be equal to those of
control units, while the three other treatments were distinguished (thin=control), and 3) I constrained all units that were burned to have equal survival rates, while survival rates on thin units were equal to those of control units (burn). Because this later model only employs one term for the treatment effect, it was possible to model an interaction between time and treatment (period*burn) without creating an over-parameterized model.
16
For all models, I constrained estimates of survival to be equal across all sites for the pre-
treatment trapping periods.
I added each of the four parameterizations of space and each of the three
parameterizations of treatment effects to the period, season and null construction of time.
I also constructed models that included only temporal variation in survival. I presumed
treatments caused differential survival between neighboring units, so I did not add terms
grouping units in space to treatment models. The only spatial term I included in
treatment models was the individual unit term. By including this term I presumed there
was within-treatment variation between the three replicates. These combinations of time,
space, and treatment in model structure constituted my a priori set of candidate models.
I used the same model structure of temporary emigration, capture, and recapture
probabilities for all models. Given the average home range of a cotton mouse is 0.5 ha
and the average life span is ≤ 5 months (Wolfe and Linzey 1977), I judged it unlikely an
animal caught during primary period i, would emigrate from the trapping area, be
unavailable for capture during period i+1, survive and return to the trapping area to be
available for capture during period i+2 (Kendall et al. 1995). Therefore, I set the
probability of an individual temporarily emigrating from the sampling grid between
primary periods to be equal to zero.
I could not calculate unique capture and recapture probabilities for each secondary period, because the size of my data set exceeded the computational capacity of program MARK. Instead, I calculated only one probability of capture and one probability of recapture for each unit for each primary period. A general, time varying model of capture probabilities such as this is appropriate for estimating population size
17
(Burnham and Anderson 2002). I modeled a trap response within primary periods by setting recapture probabilities equal to initial capture probabilities plus an additive behavioral response. I defined previous capture history with a dummy variable indicating
whether the individual had been caught in the previous primary period and used this
covariate to model a trap response across primary periods.
Inclusion of individual covariates in models requires program MARK to retrieve
values from the capture history of each individual, which can be unacceptably slow for a
large number of individuals. Therefore, I adopted a two-step model selection process.
First, I ran my a priori set of candidate models of survival without any individual
covariates. From these models I chose all those with a ΔAICc ≤ 7, added age as a
covariate with survival, added age, sex, and previous capture history as covariates with
capture probabilities, and re-ran the models. ΔAICc ≤ 7 is a conservative cut-off point
(Burnham and Anderson 2002), insuring the best approximating model was included in
the second generation of models. I drew conclusions from the results of this second
generation of models.
Recruitment models
Pradel (1996) showed that a recruitment analysis could be performed by reversing
the order of the capture histories and performing a survival analysis. His sampling design
includes only one opportunity to recapture or re-sight an individual per sampling
occasion. To perform a recruitment analysis in the robust design, which has multiple
opportunities for recapture (secondary periods) per sampling session (primary period), I
reversed the order of the primary periods while retaining the forward time order of the
secondary periods. I believe that the secondary periods are asymmetrical regarding time
18
and that proper estimation of capture probabilities over secondary periods requires a
forward time analysis. I followed the example of Nichols et al. (2002), using multistate
modeling to separate age classes into states – adults and sub-adults. The robust design
multistate models in program MARK integrate multistate modeling developed by
Brownie et al. (1993) and Hestbeck et al. (1991) with the full likelihood robust design
model of Kendall et al. (1995). The model calculates survival, initial capture probability, and recapture probability for each state, as well as the probability of transitioning from one state to the other or remaining in the same state. The closed robust design multistate model removes temporary emigration from the likelihood. I modeled differences in
capture probabilities among animals with an additive behavioral response within primary
periods and by using the Huggins estimator to include individual covariates in the
estimation of capture probabilities.
Models of seniority resembled the survival models. I implemented the period
time structure for seniority (the best structure of time for survival). I could not
implement the unit*time interaction for capture probabilities due to computational
limitations of program MARK. Therefore, I used unit as an additive effect on time. I
modeled a behavioral response to trapping between and within the primary periods as
described for survival models. I only modeled a behavioral response across primary
periods for the adult state (sub-adults have no previous capture history). I did not include
sex as an individual covariate with capture probabilities because it caused program
MARK to be unable to estimate some parameters.
Because sub-adults could not have been in the population during the previous
primary period, I set the seniority probability for sub-adults to be zero. There were not
19
enough captures of sub-adults to calculate separate capture rates for all periods and
separate transition rates for all intervals. I constrained sub-adult capture and recapture probabilities to be equal for all July primary periods and June of 2004. I estimated separate transition probabilities for each of the spring intervals; all other intervals were constrained to have the same estimate for transition probability.
I implemented treatment models in the same manner as I did with the survival
models. I replaced spatial terms used in survival modeling with unit specific measures of
the total area of bottomland hardwoods within 900 m and average distance to all
bottomland hardwoods within 900 m. Using ArcGIS, I mapped the 15 trapping grids and
all bottomland hardwoods within the study area (Figure 1). The Alabama Gap Analysis
Project provided a xx m resolution map layer of bottomland hardwoods and swamps,
created from National Wetlands Inventory maps and National Land Cover Data. I placed
a 900 m buffer around each trapping grid and within each buffer summed the total area of
bottomland hardwoods and calculated the average distance to the trapping grid from all
map cells containing bottomland hardwoods. For entry into program MARK, I
standardized these values by subtracting the mean of all values and dividing by the
standard deviation.
20
Population size and growth
For each experimental unit I calculated a time-averaged rate of population growth
(λ; Caswell 2001) for the 3-year span of the study, using the following formula:
( t −loglog NN 0 ) t λ = e
where t = 3 (time span of three years), Nt = population size from the last period of the
study (period 10), and N0 = the population size at the same time of year as period 10,
three years previous, before treatment (period 2). Estimates of population size were
taken from the selected best model of survival from program MARK.
RESULTS
Over 105,000 trap nights I captured 1,190 cotton mice 5,028 times. Fifty-six
mice died in traps (1.11 % of total captures). Out of 227 sub-adults captured, 65 (32 %)
were caught in the winter of 2003 and 102 (50 %) were caught in the winter of 2004. Out
of 67 reproductive females captured, 28 (42 %) were caught in the winter of 2003, 18 (27
%) were caught in the winter of 2004, and 15 (22 %) were caught in July 2004.
Survival models
Seven of the original 30 models had a ΔAICc ≤ 7 and were revised with individual
covariates (Table 1). All of these structured survival periodically with regard to time.
Virtually no support existed for structuring survival seasonally (0.009 summed weight for
all seasonal models) or without a time effect (0.007 summed weight for all time invariant
models).
Model period best explained the data, allowing survival to vary for each interval between primary periods but not spatially or by treatment (Table 1). The second best
21
model period + neighborhood added the 900 m spatial configuration and the third best
model period + thin=control added a treatment effect with the thin treatment being equal
to the control. Following guidelines suggested by Burnham and Anderson (2002), I
characterize model period + neighborhood and model period + thin=control as having
strong and moderate support, respectively, for consideration as the best approximating
model, given uncertainty in the information criterion. Model period was only 2.70 times
better than model period + neighborhood and 3.39 times better than model period +
thin=control. Other models had little support.
Estimates of monthly adult survival rates from the top three models ranged from
0.57 (95 % CI = 0.47, 0.66; model period + neighborhood) to 0.80 (95 % CI = 0.75,
0.84; model period + thin=control). Sub-adult survival was 9 % less than that of adults
(in model period β = -0.35, 95 % CI = -0.63, -0.07). Survival estimates for adults are
given from model period in Figure 3, illustrating variation through time. There is a
notable jump in survival in the fall after treatments were first applied, although this
model does not include a treatment effect.
Confidence intervals for all treatment and spatial effects included zero, making
survival estimates between models statistically indistinguishable. Still, treatment and
spatial terms hold information, as evidenced by competitive model AICc values and a
decrease in deviance (Table 1; Burnham and Anderson 2002). This information indicates
trends in the data that can help evaluate hypotheses. Model period + neighborhood
suggested group 1 had the highest survival rates, followed by group 3, group 2, unit 1, and finally unit 11. Units 1 and 11 were burn units. Model period + thin=control
22
suggests survival is lower on burn units and higher on thin/burn and herbicide/burn units,
relative to the thin and control units.
Initial capture probabilities from model period ranged from 0.013 (95 % CI =
0.00, 0.10) to 0.53 (95 % CI = 0.37, 0.68). Behavior, age, and sex all affected capture probabilities. Effect sizes were the virtually the same for all models, since model parameterization was the same. Cotton mice responded positively to trapping, both within and between primary periods. Recapture rate was 35 % higher than the
probability of initial capture (in model period β = 1.55, 95 % CI = 1.36, 1.73).
Individuals that had been captured during primary period i-1 were 32 % more likely to be
captured during primary period i (in model period β = 1.41, 95 % CI = 1.18, 1.63). Sub-
adults were 5 % less likely to be captured (in model period β = -0.33, 95 % CI = -0.51, -
0.15). Males were 2 % less likely to be captured (in model period β = -0.12, 95 % CI = -
0.21, -0.02).
Recruitment models
The top 11 models held 95.7 % of the AICc weight (Table 2). This can be viewed
as a 95 % confidence set of models (Burnham and Anderson 2002). Any of the top four
models could have been the one that best approximates truth, given the set of candidate
models. These models had ΔAICc < 3, held 76.8 % of the AICc weight, and had evidence
ratios that compared favorably with the top model. All of these models contained effects
for total area of source habitat within 900 m, average distance to source habitat within
900 m, and treatment. The parameterization that equates thin units to control units
(thin=control) is a more parsimonious characterization of treatment effects than handling
thin units separately (trt).
23
Seniority rates were high. Estimates from the top ranked model ranged from 0.55
(95 % CI = 0.42, 0.67) to 0.99 (95 % CI could not be calculated because the estimate was near the upper maximum likelihood boundary). Estimates of recruitment from model period show variation through time (Figure 4). Recruitment was highest in the summer and fall-spring period before treatment. After treatment, recruitment was highest in the fall, followed by the spring, and lowest in the summer.
Area of source habitat had a negative effect on seniority (in the top model β =
-0.17, 95 % CI = -0.31, -0.03). Average distance to source habitat had a positive effect on seniority (in the top model β = 0.17, 95 % CI = 0.02, 0.32). The effect of burning on recruitment was 3 % to 4 % (in the top model β = -0.51, 95 % CI = -0.87, -0.16).
Other treatment effects could not be distinguished statistically from the controls.
Regardless, the model selection process indicates the model terms for these treatments contain valuable information about recruitment. The top two models (with the thin=control parameterization) indicate that recruitment was highest on burn units, followed by herbicide/burn units, thin and control units, and finally thin/burn units. The third and fourth ranked models (with the trt parameterization) indicate the same order but with units with the thin treatment having higher recruitment than control units.
Population size and growth
Population size estimates from model period ranged from 1.22 (95 % CI = 1.01,
5.65) to 63.30 (95 % CI = 13.97, 383.9). Populations generally showed a spike during the winter trapping season. Burn units had the highest observed rate of population growth for the three-year span of the study (Figure 5). Control units had the lowest rates,
24
showing declining populations. Thin/burn units showed stable or slightly declining rates.
Thin units and herbicide/burn units showed mixed results.
DISCUSSION
My goal was to conduct a thorough assessment of cotton mouse population dynamics through intense capture-recapture sampling under the robust design. This effort, in conjunction with a replicated, manipulative experiment, has yielded reasonably strong inferences regarding the effects of ecosystem restoration efforts on cotton mice.
This is the first reported use of robust design modeling to estimate survival for adult and juvenile cotton mice, as well as seniority and recruitment estimates for cotton mice.
Treatment effects
To interpret treatment effects, I discuss recruitment rates exclusive of the effects of source habitat. I refer to estimates of survival from the third ranked survival model, which included treatment effects.
My hypothesis that thinning would not alter the habitat in a way that affects cotton mice and that habitat quality on thin sites is similar to that of controls was supported. Survival and recruitment models that treated thin units equal to control units outperformed models that treated thin separately. Results do not support the hypothesis that cotton mice would be able to locate adequate resources to survive and reproduce on thin and control units. With the exception of one thin unit, control and thin unit population growth rates were much lower than 1.0, despite the general rise in survival rates. Thin and control units had the third lowest survival and recruitment rates, indicating habitat quality on these units was poor. Perhaps the shrub layer became so
25
dense that grasses and forbs were shaded out and the mice lost a valuable source of seeds,
or the leaf litter became so deep that it was difficult for mice to forage for food.
I hypothesized that burning would improve habitat for cotton mice by increasing local availability of resources, with the thin/burn treatment causing the best improvement.
This hypothesis was met with interestingly mixed results. The thin/burn treatment may have created sources. This treatment had the highest survival rates, the lowest recruitment rates, and population growth rates close to one. A source is expected to have a stationary population growth rate over several generations as resident individuals live long and only enough individuals are recruited to the population to replace those that die;
excess individuals are exported (Pulliam 1988). One argument against this conclusion is
that if the thin/burn treatment improved habitat, raising the carrying capacity, the size of
the populations should have increased from pre-treatment levels. The population growth
rate on two of these sites was 0.94, which could be viewed as too low to describe a
source. The population growth rate on the third site was 1.04. These calculations were
made from estimates of population size, which have an associated variance, so estimates
of population growth rate are approximations. Real values could be higher or lower.
The herbicide/burn treatment also appears to have improved habitat for cotton
mice. Two of these sites had growing populations. The herbicide/burn units had the
second highest survival rates and recruitment rates, improved over control units.
Unlike the thin/burn treatment, the burn treatment appears to have created sinks.
These units had the highest population growth rates and highest recruitment rates but the
lowest survival rates. These rates are characteristic of a sink wherein turnover of
individuals is high. Sinks can have population sizes much higher than that of sources
26
(Pulliam 1988) and a newly created sink could exhibit a rapidly growing population as animals displaced from sources spill into the area.
One possible explanation for the difference between the thin/burn and burn treatment is that the combination of thinning and burning had a synergistic effect.
Opening the shrub layer by fire and the canopy by thinning allowed light to reach the forest floor so the beneficial effects of fire could be realized. On burn sites, where thinning was absent, it may be that not enough light reached the forest floor for the beneficial effects of fire to be realized, leaving only a burned-out understory that provided few resources for mice. This does not explain how the herbicide/burn treatment improved habitat. An alternative explanation is that the combination treatments did a better job of opening the shrub layer, allowing the herbaceous layer to flourish, whereas burning alone was not enough to sufficiently open the shrub layer to benefit the herbaceous layer.
Results indicate treatments had a stronger effect on recruitment than survival.
The top ranked survival model included no variation in survival estimates. The thin=control parameterization of survival ranked third and estimated treatment effects included zero in the 95 % confidence interval. This implies variation in survival rates between treatments was relatively small. Conversely, there was enough variation in recruitment between treatments that treatment models performed strongly in the recruitment analysis and zero was not included in the estimate of the 95 % confidence interval for the effect of the burn treatment. Even small differences in survival between treatments could cause noticeable changes in the population growth rate, as evidenced by high seniority estimates (Nichols et al. 2000).
27
Source habitat effects
Results provide circumstantial support for the hypothesis that bottomland
hardwood forests and swamps serve as sources for cotton mice. Results supported my
hypothesis that bottomland hardwoods close to experimental units would provide immigrants to those units. Recruitment increased with an increase in the total area of bottomland hardwoods within 900 m and decreased as the average distance to bottomland hardwoods within 900 m increased. Models that combined both source habitat and treatment effects ranked highest in the recruitment analysis. This is the first use of landscape variables in capture-recapture modeling to infer the immigration of animals from source habitat outside the study area.
This difference in immigration between experimental units caused by availability
of source habitat did not obscure treatment effects on population size. By ranking units
in descending order of population growth rate, a clear pattern of treatment effect emerges
(Figure 5). The three burn sites had the highest rates, thin/burn sites had stable (or nearly stable) rates, and control sites had the lowest rates. This pattern is coincident with estimates of recruitment and survival from models including treatment terms. The pattern was not perfect as one herbicide/burn unit and one thin unit were not in the expected order. Neither this departure from expected order, nor the rankings of units within treatments were explained by availability of source habitat.
Spatial effects on survival
The rank of units by population growth rate within the thin/burn, thin, and control
treatments is best explained by the period + neighborhood survival model. The strong contribution to population growth rate by survival apparently overshadowed differences
28
in recruitment between units caused by immigration from source habitat. There is no
apparent explanation for the order of units with the burn and herbicide/burn treatments.
Again, estimates of population growth rate do not account for variation in population size
estimates, so failure to achieve an exact match between observed population growth rates
and model predictions is not surprising.
Model selection did not favor models that included the unit or locale spatial terms, indicating populations could not be defined over these smaller areas. Defining the
cotton mouse population requires some resolution greater than the 570 m radius used for
buffers to define the locale model. This resolution may be large, as the top ranked
survival model contained no terms of spatial segregation.
Habitat quality, vital rates, and spatial scale
Defining habitat quality requires delimiting the boundaries of the habitat and linking vital rates to the delimited area (Garshelis 2000). These two objectives are notoriously difficult to obtain. Habitat is usually defined arbitrarily from a human
perspective and vital rates are rarely measured in studies of habitat quality. I constructed
models that linked vital rates to habitat defined at various scales, based upon insights
from cotton mouse biology. Model selection results show that vital rates operated over
multiple scales. The top three survival models each defined habitat on a different scale.
The top model (period) defined habitat as the entire study area. The second ranked
model (period + neighborhood) defined habitat at an intermediate scale. The third ranked
model (period + thin=control) defined habitat by treatments, which were applied to
small, 15 ha stands. Recruitment also varied at different scales due to quality of habitat
both outside and within experimental units. Some of these results are surprising. It is
29
easy to understand how the survival of a mouse that lives on a 0.5 ha home range is
determined by local habitat quality. It is not as intuitive to think about survival rates for
the population as being determined by habitat defined over a large area.
Defining the scale at which vital rates operate provides a better idea of the
appropriate size area for restoration projects. In my study, treatments applied to 15 ha
stands were probably not large enough to substantially effect the population; rather they
caused localized departures in vital rates within the larger population. Information about
the size of populations and the space they require to function is also very important for
determining the necessary size of wildlife refuges. Community membership within a
management or conservation area may be determined, in part, by the available pool of
species living outside the area but within dispersal range.
Management implications
Small mammals play a key role in ecosystem function (Grant and French 1980,
Brown and Heske 1990, Talmon et al. 2003) and therefore judging the success of ecosystem restoration efforts should include consideration of the effects on the small mammal community. Assuming cotton mice represented the small mammal community, combining fire with thinning or herbicide appeared to have the greatest impact, and therefore may be the quickest method for restoring the ecosystem. Had treatments been carried out over a larger area, treatment effects might have been more substantial.
Applying restoration efforts over a large area would mimic the fires that once burned over great distances in the longleaf pine ecosystem.
Results indicated that cotton mice populations in the upland were supported, at least in part, by immigration from sources outside the management area. Maintaining
30
refuges of source habitat can help mitigate the effects of forest management on the small
mammal community. Managers should be cognizant of how their actions might attract immigrants from outside the management area.
Although a treatment effect was apparent in population growth rates,
interpretation of these rates alone could lead to false conclusions. Without information
about survival, the high estimates of recruitment and population growth on the burn units
could give the impression that the treatment had a positive effect. Managing animal
communities requires assessment of changes in population demographics as they respond
to habitat alteration at a local scale, within the context of the surrounding landscape. An
accurate assessment of the effects of restoration actions on an animal population requires
measurement of survival and reproduction within the restoration area and animal
movement to and from the restoration area.
ACKNOWLEDGEMENTS
I greatly appreciate the assistance of S. Converse with program MARK and model
development. J. Nichols and G. White provided helpful suggestions and insights. I am
grateful to all the field technicians that helped with the small mammal trapping. J.
Christian provided administrative support. K. Kleiner contributed the GIS map layer of
bottomland hardwoods and swamps. This is contribution number XX of the National
Fire and Fire Surrogate Research (FFS) Project. This research was funded by the USDA
Forest Service through the National Fire Plan. Although the author received no direct
funding for this research from the U.S. Joint Fire Science Program (JFSP), it was greatly
facilitated by the JFSP support of existing FFS project sites.
31
LITERATURE CITED
Akaike, H. 1973. Information theory and an extension of the maximum likelihood
principle. Pages 267-281 in B. N. Petran and F. Csaaki, editors. Second
international symposium of information theory. Akadeemiai Kiadi, Budapest,
Hungary.
Boyer, W. D. 1964. Longleaf seed losses to animals on burned seedbeds. United States
Department of Agriculture, Forest Service, Southern Forestry Experimental
Station, Research Note SO-6:1-3.
Brockway, D. G. and K. W. Outcalt. 2000. Restoring longleaf pine wiregrass ecosystems:
hexazinone application enhances effects of prescribed fire. Forest Ecology and
Management 137:121-138.
Brown, J. H. and E. J. Heske. 1990. Control of a desert-grassland transition by a keystone
rodent guild. Science 250:1705-1707.
Brownie, C., J. E. Hines, J. D. Nichols, K. H. Pollock and J. B. Hestbeck. 1993. Capture-
recapture studies for multiple strata including non-Markovian transition
probabilities. Biometrics 49:1173-1187.
Burnham, K. P. and D. R. Anderson. 2002. Model selection and multimodel inference: a
practical information-theoretic approach. Springer-Verlag, New York, New York,
USA.
Caswell, H. 2001. Matrix population models. Second Edition. Sinauer Associates,
Sunderland, Massachusetts, USA.
32
Cooch, E. and G. C. White. 2004. Using MARK – a gentle introduction. Third edition.
United States Geological Survey, New York and Colorado Cooperative Fish and
Wildlife Research Units. Ithaca, New York, USA.
ESRI. 2004. ArcGIS 8.3. Environmental Systems Research, Inc., Redlands, California,
USA.
Fischer, R. A. and N. R. Holler. 1990. Habitat use and relative abundance of gray
squirrels in southern Alabama. Journal of Wildlife Management 55:52-59.
Frost, C. C. 1993. Four centuries of changing landscape patterns in the longleaf pine
ecosystem. Pages. 17-43 in S. M. Hermann, K. Ross, and K. Gainey, editors. The
longleaf pine ecosystem: ecology, restoration, and management. Proceedings of
the 18th Tall Timbers fire ecology conference. Tall Timbers Research, Inc.,
Tallahassee, Florida, USA.
Garshelis, D. L. 2000. Delusions of habitat evaluation. Pages 111-164 in L. Boitani and
T. K. Fuller, editors. Research techniques in animal ecology. Columbia University
Press, New York, New York, USA.
Grant, W. E. and N. R. French. 1980. Evaluation of the role of small mammals in
grassland ecosystems: a modeling approach. Ecological Modeling 8:15-37.
Hatchell, G. E. 1964. Small mammal species and populations in the loblolly-shortleaf
pine forest type of Louisiana. United States Department of Agriculture, Forest
Service, Southern Forestry Experimental Station, Research Paper SO-10:1-12.
Hestbeck, J., J. D. Nichols, and R. A. Malecki. 1991. Estimates of movement and site
fidelity using mark-resight data of wintering Canada geese. Ecology 72:523-533.
33
Hofstetter, R. H. 1973. An ecological study of the effects of fire on the wet prairie,
sawgrass glades, and pineland communities of south Florida. Pages 95-116 in
Effects of fire in the ecosystem. Final Report Part 2. USDI National Park Service,
EVER-N-48. NTIS No. PB264463.
Huggins, R. M. 1989. On the statistical analysis of capture experiments. Biometrika
76:133-140.
Huggins, R. M. 1991. Some practical aspects of a conditional likelihood approach to
capture experiments. Biometrics 47:725-732.
Kendall, W. L., K. H. Pollock, and C. Brownie. 1995. A likelihood-based approach to
capture-recapture estimation of demographic parameters under the robust design.
Ecology 78:563-578.
Landers, J.L., N.A. Byrd, and R. Komarek. 1990. A holistic approach to managing
longleaf pine communities. Pages 135-167 in R.M. Farrar Jr. editor. Proceedings
of the symposium on the management of longleaf pine. United States Department
of Agriculture, Forest Service, General Technical Report SO-75:1-293.
Layne, J. N. 1974. Ecology of small mammals in a flatwoods habitat in north central
Florida, with special emphasis on the cotton rat (Sigmodon hispidus). American
Museum Novitates 2544:1-48.
Nichols, J. D., J. E. Hines, J. D. Lebreton, and R. Pradel. 2000. Estimation of
contributions to population growth: a reverse-time capture-recapture approach.
Ecology 81:3362-3376.
34
Outcalt, K.W. 2002. Study plan: fire and fire surrogate treatments in the Gulf Coastal
Plain Solon Dixon Forestry and Education Center, Alabama. United States
Department of Agriculture, Forest Service, Southern Research Station, Athens,
GA, USA.
Peet, R. K. and D. J. Allard. 1993. Longleaf pine vegetation of the southern Atlantic and
eastern Gulf coast regions: A preliminary classification. Pages 45-82 in S. M.
Hermann, K. Ross, and K. Gainey editors. The longleaf pine ecosystem: ecology,
restoration, and management. Proceedings of the 18th Tall Timbers fire ecology
conference. Tall Timbers Research, Inc., Tallahassee, Florida, USA.
Platt, W. J. 1999. Southern pine savannas. Pages 7-22 in R. C. Anderson, J. S. Fralish,
and J. M. Baskin, editors. Savannas, barrens, and rock outcrop plant communities
of North America. Cambridge University Press, Cambridge, UK.
Pollock, K. H. 1982. A capture-recapture sampling design robust to unequal catchability.
Journal of wildlife management 46:752-757.
Pollock, K. H., D. L. Solomon, and D. S. Robson. 1974. Tests for mortality and
recruitment in a K-sample tag-recapture experiment. Biometrics 30:77-87.
Pournelle, G. H. 1952. Reproduction and early post-natal development of the cotton
mouse, Peromyscus gossypinus gossypinus. Journal of Mammalogy 33:1-20.
Pradel, R. 1996. Utilization of capture-mark-recapture for the study of recruitment and
population growth rate. Biometrics 52:703-709.
Pulliam, H. R. 1988. Sources, sinks, and population regulation. The American Naturalist
132:652-661.
35
Shadowen, H. E. 1963. A live-trap study of small mammals in Louisiana. Journal of
Mammalogy 44:103-108.
Stransky, J. J. and R. F. Harlow. 1981. Effects of fire on deer habitat in the Southeast.
Pages 135-142 in G.W. Wood, editor. Prescribed fire and wildlife in southern
forests: proceedings of the symposium. The Belle W. Baruch Forest Science
Institute, Clemson University, Georgetown, South Carolina, USA.
Tallmon, D. A., E. S. Jules, N. J. Radke, and L. S. Mills. 2003. Of mice and men and
trillium: cascading effects of forest fragmentation. Ecological Applications
13:1193-1203.
Van Horne, B. 1983. Density as a misleading indicator of habitat quality. Journal of
Wildlife Management 47:893-901.
Van Lear, D. H and R. F. Harlow. 2000. Fire in the eastern United States: Influence on
wildlife habitat. Pages 2-10 in W. M. Ford, K. R. Russell, and C. E. Moorman,
editors. The role of fire in non-game wildlife management and community
restoration: traditional uses and new direction. Proceedings of a special workshop.
United States Department of Agriculture, Forest Service, Northeastern Research
Station, General Technical Report NE-288:1-145.
Weatherspoon, C. P. and J. McIver. 2000. A national study of the consequences of fire
and fire surrogate treatments. Proposal to the Joint Fire Science Program. United
States Department of Agriculture, Forest Service, Pacific South West Research
Station.
White, G. C. and K. P. Burnham. 1999. Program MARK: survival estimation from
populations of marked animals. Bird Study 46 Supplement:120-138.
36
Williams, B. K., J. D. Nichols, and M. J. Conroy. 2002. Analysis and management of
animal populations. Academic Press, New York, New York, USA.
Wolfe, J. L. and A. V. Linzey. 1977. Peromyscus gossypinus. Mammalian Species 70:1-
5.
37
TABLES AND FIGURES
Table 1. Summary of model selection results for survival of cotton mice in Covington and Escambia Counties, Alabama, 2001-2004. Models are ranked in ascending order by
Akaike’s Information Criterion, adjusted for small sample size (AICc). Top models all estimated a unique survival rate for each of nine intervals between trapping periods
(period). Neighborhood and locale represent two spatial aggregations of experimental units. The neighborhood configuration was constructed by placing 900 m buffers around experimental units, while the locale configuration was constructed with a maximum 570 m buffer. Treatments were modeled in 1 of 3 ways: 1) each treatment – thin, burn, thin/burn, herbicide/burn, and control – was estimated to have unique effect (trt), 2) thin units were estimated to have the same estimates of survival as control units; all other treatments had a unique effect (thin=control), 3) all units that were burned were constrained to have the same estimate of survival; thin units were constrained to have the same estimates of survival as control units (burn).
38
3 AIC Evidence Cumulative4 1 c 2 Model AICc ΔAICc Weight K Deviance Ratio Weights
period 17496.15 0.00 0.56 150 17186.76 0.56 period + neighborhood 17498.14 1.99 0.21 154 17180.23 2.70 0.76 period + thin=control 17498.63 2.48 0.16 153 17182.86 3.46 0.92 period + trt 17500.85 4.70 0.05 154 17182.95 10.50 0.97 period + locale 17502.72 6.57 0.02 155 17182.69 26.75 1.00 period * burn 17505.62 9.47 0.00 157 17181.32 114.14 1.00 period + burn 17539.27 43.12 0.00 151 17227.75 2.31E+09 1.00
1 Akaike’s Information Criterion corrected for small sample size
2 Number of parameters
3 Likelihood of the top ranked model versus the competing model (e.g., the top model is
2.7 times more likely to be the model that best approximates truth than the second
ranked model).
4 Sum of the AICc weights from the competing model and each higher ranked model.
39
Table 2. Model selection results for recruitment of cotton mice in Covington and
Escambia Counties, Alabama, 2001-2004. Models are ranked in ascending order by
Akaike’s Information Criterion, adjusted for small sample size (AICc). Treatments were modeled in 1 of 3 ways: 1) each treatment – thin, burn, thin/burn, herbicide/burn, and control – was estimated to have unique effect (trt), 2) thin units were estimated to have the same estimates of survival as control units; all other treatments had a unique effect
(thin=control), 3) all units that were burned were constrained to have the same estimate of survival; thin units were constrained to have the same estimates of survival as control units (burn). Area = total area of bottomland hardwoods within 900 m of each experimental unit. Avg dist = average distance to all 30 m map cells containing bottomland hardwoods within 900 m of each experimental unit.
40
3 AIC Evidence Cumulative4 1 c 2 Model AICc Δ AICc Weight K Deviance Ratio Weights
area + avg dist + thin=control 17787.17 0.00 0.33 64 17657.50 0.33
area * avg dist + thin=control 17788.10 0.92 0.21 65 17656.37 1.59 0.54
area + avg dist + trt 17788.91 1.74 0.14 65 17657.18 2.38 0.68
area * avg dist + trt 17789.89 2.72 0.09 66 17656.11 3.90 0.77
avg dist + thin=control 17790.92 3.75 0.05 63 17663.30 6.52 0.82
area * avg dist 17791.77 4.59 0.03 60 17670.29 9.94 0.85
thin=control 17791.87 4.69 0.03 62 17666.29 10.46 0.88
area * avg dist 17792.25 5.08 0.03 62 17666.68 12.67 0.91
avg dist + trt 17792.86 5.68 0.02 64 17663.18 17.15 0.93
period * burn 17793.51 6.33 0.01 61 17669.98 23.73 0.94
trt 17793.61 6.44 0.01 63 17665.99 25.00 0.96
1 Akaike’s Information Criterion corrected for small sample size
2 Number of parameters
3 Likelihood of the top ranked model versus the competing model (e.g., the top model is
2.7 times more likely to be the model that best approximates truth than the second
ranked model)
4 Sum of the AICc weights from the competing model and each higher ranked model.
41
Figure 1. Map of Auburn University’s Solon Dixon Forestry Education Center, site of
the Fire/Fire Surrogate Study’s Gulf Coastal Plain location, showing experimental design,
900 m buffers for spatial segregation of units, and location of bottomland hardwoods and
swamps.
Figure 2. Treatment and trapping schedule for a replicated experiment and capture-
recapture study of the effects of fire and fire alternatives on small mammals in longleaf
pine stands, Covington and Escambia Counties, Alabama, 2001-2004.
Figure 3. Monthly survival rates and 95% confidence intervals, estimated for each
interval between primary trapping periods (model period), for cotton mice in Conecuh
and Escambia Counties, Alabama, 2001 – 2004.
Figure 4. Monthly recruitment rates and 95% confidence intervals, estimated for each interval between primary trapping periods, for cotton mice in Conecuh and Escambia
Counties, Alabama, 2001 – 2004. Recruitment in the summer of 2003 was almost zero.
Confidence intervals for an estimate approaching the maximum likelihood boundary cannot be calculated accurately.
Figure 5. Population growth rates for a three-year span from 2001 to 2004, ranked in descending order, for populations of cotton mice living on experimental units treated for habitat restoration, in Covington and Escambia Counties, Alabama.
42
Figure 1.
43
Figure 2.
Primary period
12 34 5 67 8 910
Burn herbicide 2nd burn Thin Burn Herbicide units 2nd burn unit 11
1 3 01 02 03 04 0 002 00 004 0 200 20 2 20 2 2 y 2002 ul 2003 n 2 ul 2004 un 2 Jul a un 2002Jul ep ay J ay J J J S Feb 2003 Jun Feb 2004 Ju r-M r-M r-M p Feb-AprAp 2002 A Ap
44
Figure 3.
0.9
0.8
0.7
0.6 Monthly survival rate Monthly
0.5
0.4 2 003 004 2001 200 2002 2 2003 2 g 2003 g 2004 nter nter prin prin S S ummer ummer - Wi ummer - Wi ummer S S l S l S al al F F
Fall 2001 - Spring 2002
45
Figure 4.
0.6
0.5
0.4
0.3
0.2 Monthly recruitment rate recruitment Monthly
0.1
0.0 2 3 001 2002 r 200 2003 r 200 2004 mer 2 ring 2002mer ring 2003mer ring 2004mer p nte p nte p um um Wi S um Wi S um S S - S - S Fall Fall
Fall 2001 - S
46
Figure 5.
1.6 Control Burn 1.4 Thin Herbicide/burn Thin/burn 1.2
1.0
0.8 Population growth rate
0.6
0.4
123456789101112131415
Rank
47
RH: Small mammals of longleaf pine
RESTORING THE SMALL MAMMAL COMMUNITY OF A LONGLEAF PINE
ECOSYSTEM
Abstract: Small mammals play a key role in ecosystem function and should be
considered in any evaluation of the effectiveness of ecosystem restoration efforts. I
participated in a replicated, manipulative experiment investigating the use of fire and fire
alternatives to restore a longleaf pine ecosystem, formerly one of the most expansive and
biologically diverse ecosystems of North America. I tracked changes in the small
mammal community and species’ demographics after application of restoration
treatments. To define restoration goals for the small mammal community, I sampled a
nearby longleaf pine savanna that was maintained by regular fires and resembled the
historic condition. I found the reference small mammal community and the community
to be restored differed in membership and structure. No changes in community structure
were observed after treatment, but changes at the population level were evident; fire
affected golden mouse (Ochrotomys nuttalli) survival negatively. Spring and summer
survival rates for cotton mice (Peromyscus gossypinus) in the savanna were higher than
those of cotton mice living in habitat being restored. Restoration of the small mammal
community may be slow and several applications of fire may be required. Spring and summer survival rates for cotton mice in the savanna were also higher than those reported
for the Key Largo cotton mouse (P. g. allapaticola) living in hammocks on Key Largo,
48
Florida, which contradicts the belief that hammocks provide optimal cotton mouse
habitat. Optimal habitat for cotton mice may be a composite of longleaf pine uplands and
hardwood bottomlands.
Key words: capture-recapture, demography, ecosystem restoration, fire, longleaf pine,
Ochrotomys nuttalli, Peromyscus gossypinus, robust design, small mammals, survival
Habitat restoration projects require measurable goals so that managers can
determine whether restoration actions are having the desired effect and judge when
restoration has been successful. Often restoration efforts are focused only on the floral
community, but wildlife can play a key role in shaping the ecosystem (e.g., a guild of
kangaroo mice (Dipodomys spp.) in the Chihuahuan Desert [Brown and Heske 1990],
elephants (Loxodonta africana) in Africa [Dublin et al. 1990], and wolves (Canis lupus)
in North America [Ripple and Beschta 2004]). A holistic approach to ecosystem
restoration should define desired changes in both the floral and faunal communities.
Restoring fire-maintained ecosystems is now a concern in the United States.
Decades of fire suppression policy have led to an annual epidemic of catastrophic forest fires across the nation. The Fire/Fire Surrogate (FFS) study, a collaborative effort
between the United States Department of the Interior and the United States Department
of Agriculture (Weatherspoon and McIver 2000) was designed to determine the most
ecologically and economically efficient methods to reduce fuel loads and restore the
nation’s fire-maintained ecosystems. I measured the effects of experimental treatments
on small mammals at the Gulf Coastal Plain site of the FFS study, where longleaf pine
(Pinus palustris) savanna was the fire-maintained ecosystem to be restored.
49
Longleaf pine was once one of the most extensive forest types in North America
(Landers and Boyer 1999), stretching from southeastern Virginia to eastern Texas (Frost
1993). The longleaf pine ecosystem supports a high diversity of plant species, making it
one of the most species-rich communities outside of the tropics (Peet and Allard 1993).
It is also home to the threatened gopher tortoise (Gopherus polyphemus) and indigo snake
(Drymarchon corais) and the endangered red cockaded woodpecker (Picoides borealis).
Today longleaf pine occupies < 3 % of its original range (Frost 1993). The loss of this expansive, unique, and species-rich ecosystem is of great ecological concern.
Early explorers to North America described expansive savannas featuring an open grass/forb understory and sparsely distributed longleaf pines forming open canopies (Platt
1999). These savannas evolved under a process of frequent fires ignited by the lightning
strikes of spring storms (Platt 1999). Fires would burn at low intensity, sometimes over
great distance, through the continuous grass understory (Frost 1993; Peet and Allard
1993). In the absence of fire, the fire tolerant species of longleaf pine savannas lose their
competitive advantage over fire-intolerant species. Hardwoods eventually gain
dominance in the canopy, shading out the pines, and a dense shrub understory develops,
shading out the natural grasses, herbs, and longleaf pine saplings (Landers and Boyer
1999). Without fire, fuel loads build in the woody understory, which can lead to high
intensity, crown-scorching wildfires (Peet and Allard 1993).
Information about the small mammal community of the longleaf pine ecosystem
is sparse. No demographic information on small mammals living in longleaf pine stands exists. Boyer (1964) is the only researcher to document the small mammals of longleaf pine stands in Alabama. Out of 46 captures, he caught 25 oldfield mice (Peromyscus
50
polionotus), 14 cotton mice (P. gossypinus), 4 least shrews (Cryptotis parva), 2 eastern harvest mice (Reithrodontomys humulis), and 1 golden mouse (Ochrotomys nuttalli).
Few studies have examined how fire affects the small mammals of longleaf pine.
Komarek (1939) reported changes in species present and their abundance in an experimental area in Grady County, Georgia, where fire was suppressed. Over 4 years, the understory changed from broomsedge and herbaceous ground cover to low bushes and tangles of vines. As the vegetation changed, the cotton rat (Sigmodon hispidus) population decreased while the cotton mouse and golden mouse population increased.
Komarek called this change in the small mammal community a mammalian succession that was correlated with plant succession. He suggested that a mammalian climax community exists for each plant climax community.
Arata (1959) and Layne (1974) each conducted a study of burned areas near
Gainesville, Florida. Arata (1959) studied an area covered with a dense understory of grasses and forbs dominated by turkey oak with few longleaf pine trees interspersed.
This area had been burned in the last 3 or 4 years, which probably accounts for the open grass/forb understory. He found no change in the oldfield mouse or Florida mouse (P. floridanus) populations after fire, but cotton rats left the area. He felt that a low number of cotton rats in areas surrounding the burned site explained their failure to reoccupy the site once vegetation had returned to a density suitable for cotton rat habitation.
Layne (1974) studied a stand of slash pine (Pinus elliottii) and longleaf pine with a saw palmetto (Serenoa repens) and shrub understory before and after it was burned for the first time in 10 years. He showed a rapid increase in cotton mice after the fire. He also showed a fluctuation in the cotton rat population. Cotton rats left a burned area in
51
conjunction with the nearly complete loss of ground cover, but returned - in greater
numbers than in the unburned area - after ground cover had grown back. The return of
eastern harvest mice and cotton rats appeared to be related to the redevelopment of
ground cover.
These studies were observational in nature and conclusions were based upon the
numbers of individuals captured. Counting individuals does not account for variation in
capture probabilities over time and space and therefore counts can be biased indices of
population size (Williams et al. 2002). This may lead to inappropriate comparisons
between sites. No study of fire in southern pinelands captured golden mice in numbers
sufficient to suggest a response.
Many studies have examined the effects of habitat manipulation on small mammals, including the use of fire, herbicide, and mechanical removal of vegetation
(e.g., Komarek 1939, Ahlgren 1966, Beck and Vogl 1972, Sullivan and Sullivan 1982,
McMurry et al. 1996, Masters et al. 1998). While the specific results of these studies are
as varied as the habitat types studied and the habitat affinities of the species living in
them, the result common to most of these studies is that changes in the three-dimensional
structure and content of the plant community cause changes in the structure and content
of the small mammal community.
The FFS study at the Gulf Coastal Plain site was the first replicated, manipulative
experiment to investigate the effects on small mammals of efforts to restore a longleaf
pine ecosystem. The FFS study employed 4 treatments representing options for forest
restoration: 1) control - no action was taken; 2) burn - prescribed fire was used to reduce
fuel levels and maintain a frequent fire regime; 3) thin - trees were removed to reduce
52
tree density; 4) thin/burn - thinning was followed by prescribed fire. All treatments were
replicated 3 times. These treatments were intended to result in a common desired future
condition (DFC) - a forest stand structure and fuel level such that 80% of the trees would be able to survive a wildfire in all but the most unfavorable weather conditions.
The Gulf Coastal Plain site was located at Auburn University’s Solon Dixon
Forestry Education Center (Dixon Center) in south-central Alabama. The Dixon Center
includes 2,130 ha of managed pine forestland. Sites selected for experimental units were
longleaf pine stands in which the natural process of fire has been interrupted; hardwoods
have grown into the canopy alongside the dominate longleaf pine and at the time of
selection had a dense shrub understory dominated by yaupon (Ilex vomitoria). The Study
Plan for the Gulf Coastal Plain site further defines the DFC as a stand structure like that
of longleaf pine savannas: a canopy dominated by longleaf pine trees and an understory of grasses and forbs with some shrubs (Outcalt 2002). Herbicide is commonly used in
pine plantations of the southeast to reduce understory vegetation that would compete with
pine trees, and could help achieve the understory of grasses and forbs specified by the
DFC. At the Gulf Coastal Plain site a 5th treatment was added to the 4 FFS treatments:
herbicide/burn - herbicide was applied to the shrubby understory followed by prescribed
fire.
To determine the DFC for the small mammal community, I expanded the study to
include Compartment 58 of Conecuh National Forest (Conecuh NF), located near the
Dixon Center. This site had a history of being burned during the spring every three years and resembled natural longleaf pine savannas; it was almost exclusively longleaf pine with a grass/forb understory. I used this site as a reference site or example of the small
53
mammal community found in local longleaf pine savannas. This allowed me to
determine if the small mammal community living in the hardwood/pine woodlands at the
Dixon Center differed from the community living in the pine savanna site of Conecuh
NF, and track the progress of restoration towards the DFC.
To evaluate the difference in population demographics of species between
treatments and between woodland and savanna, I conducted a capture-recapture study under Pollock’s robust design (Pollock 1982, Kendall et al. 1995). I used the robust design in program MARK (White and Burnham 1999) to estimate abundance and survival rates for cotton mice and golden mice. The robust design provides the most precise estimates of capture probabilities available and thereby provides superior estimates of abundance and survival (Williams et al. 2002). I compared estimates for cotton mice in Conecuh NF to estimates for cotton mice at the Dixon Center to determine if there was a difference between the 2 locations. To determine whether restoration treatments affected survival of golden mice, I used program MARK to compare competing models of survival with and without treatment effects.
I expected the experimental treatments would cause changes in the plant
community that would affect small mammals. Fire has several advantageous effects on
vegetation. Soft mast and seed production increases after fire (Van Lear and Harlow
2000), grasses and forbs expand coverage of the forest floor (Brockway and Outcalt
2000), and resprouting vegetation is more nutritious and palatable (Stransky and Harlow
1981). Due to these beneficial aspects, I expected the habitat of Conecuh NF would be of
higher quality for cotton mice than habitat at the Dixon Center. Golden mice prefer
dense brush or thickets and are often found in association with greenbriar (Smilax spp.)
54
and honeysuckle (Lonicera spp.; Linzey and Packard 1977). I expected treatments would largely remove these habitat components, reducing habitat quality for golden mice.
Cotton rats are closely associated with grassy habitats (Cameron and Spencer 1981), so I expected treatments to create favorable habitat for cotton rats.
I hypothesized longleaf pine stands maintained by frequent spring fires would support a structurally and compositionally different small mammal community than longleaf pine stands in which the natural process of fire has been interrupted. Since fire has beneficial effects on vegetation for cotton mice, savanna habitat should be of higher quality for cotton mice than woodlands. As treatments involving fire begin to shift the plant community from woodland to savanna, habitat quality for each small mammal species should change, altering each species’ population demographics and causing a subsequent change in the community structure. Fire should degrade habitat quality for golden mice, but create favorable habitat for cotton rats. The structure of the small mammal community should shift from species preferring woodlands to species preferring savannas. Thinning alone should not substantially alter habitat quality for small mammals. Given my hypotheses, I predicted that: 1) captures in Conecuh NF would include small mammals with a preference for savanna habitat, 2) captures at the Dixon
Center would include species with a preference for woodland habitat, 3) golden mouse survival and abundance would decline on burn, thin/burn, and herbicide/burn sites, 4) cotton mouse survival and abundance would be higher in Conecuh NF than at the Dixon
Center, and 5) cotton rats would immigrate into units that were burned.
55
MATERIALS AND METHODS
Study area.⎯The Dixon Center is located in Covington and Escambia counties
in the Gulf Coastal Plain of south-central Alabama (31°9’N, 086°42’W). Summers in
this region are hot and humid (often 35°C and 95% humidity), while winters are mild.
Average annual precipitation is 14.22 cm (Outcalt 2002). Most precipitation occurs
during late winter through spring and during afternoon showers in the summer. The
terrain is relatively flat with rolling hills and ranges in elevation from 30 m to 100 m
(Outcalt 2002).
Longleaf pine dominates much of the managed forest of the Dixon Center, but
loblolly, slash, shortleaf (P. echinata), and spruce pine (P. glabra) are also abundant
(Fischer and Holler 1990). Hardwoods can be found interspersed throughout the
overstory and midstory in mixed pine-hardwood stands and concentrated along
bottomland streams. Laurel oak (Q. laurifolia), white oak (Q. alba), and water oak (Q.
nigra) are most numerous, while American beech (Fagus grandifolia), sweetgum
(Liquidambar styraciflua), hickory (Carya spp.), maple (Acer spp.), ash (Fraxinus spp.), and elm (Ulmus spp.) also occur (Fischer and Holler 1990). The understory at the Dixon
Center is dominated by yaupon, along with lesser amounts of blueberries (Vaccinium spp.) and gallberry (I. glabra; Outcalt 2002). In places yaupon forms shrubby thickets that are quite dense.
Compartment 58 of Conecuh NF lies approximately 14.5 km south of the Dixon
Center (31.03° N, 086.64° W). This 155 ha stand was recently burned in the spring of
1996, 1999, and 2002. The overstory is exclusively longleaf pine. No midstory exists
and the understory consists of grasses and forbs.
56
Experimental design.⎯The experiment at the Dixon center was laid out in a
randomized block design. Three blocks were arranged by similar soil type and general
location. Each treatment and the control were replicated 3 times, for a total of 15
experimental units. One replicate of each treatment and control was randomly assigned
to an experimental unit in each block. Experimental units were 15.2 ha stands of mixed
hardwoods and pine with a dense shrub understory. Minimum distance between
experimental units was 156 m. I established 3 additional trapping grids in Compartment
58. The grids were in a line, the outside grids being 200 m and 280 m distant from the middle grid.
Treatment and trapping schedule.⎯All thinning took place between February
and April 2002 (Fig. 1). Burn and thin/burn sites were burned for the first time in April
and May 2002. Herbicide was applied in September 2002. Herbicide/burn sites were
burned in the spring of 2003. A second burning of the burn and thin/burn sites was
conducted in the spring of 2004. Due to weather conditions, one burn site was not burned
a second time until July, after the first trapping period of the summer had taken place.
Trapping followed Pollock’s (1982) robust design. I conducted ten trapping sessions
(primary periods), each comprising seven consecutive nights of trapping (secondary
periods; Fig. 1). I trapped on each experimental unit twice during the summer for the 4-
year duration of the study, with approximately one month between the two summer
periods. I began winter trapping in February 2003 to gather more demographic data during the breeding season. I began trapping in Conecuh NF with the second primary period of the summer of 2002.
57
Small mammal trapping.⎯I sampled the small mammal community using
standard capture-recapture methods. I installed a 0.81 ha trapping grid on each unit,
locating most trapping grids in the center of each experimental unit. In some cases I
opted to offset grids so as to avoid dirt roads. Each grid had 100 trapping stations,
located at the intersection of 10 rows and 10 columns placed at 10 m intervals. I placed a
single Sherman live trap, baited with sunflower seeds, at each trap station. I checked
traps each morning. During the summer, I closed traps during the heat of the day and re-
opened and baited them in the evening. During the winter, I placed batting in each trap for nesting material. Due to time constraints imposed by short winter days, I re-baited
traps and left them open during the day.
I gave each newly captured animal a uniquely numbered ear tag (monel ear tags
from Western Tag Co., Salt Lake City, Utah, USA). For every capture I recorded
standard morphological measures and individual identification number. I released
animals at the point of capture. This protocol was approved by Auburn University’s
Institutional Animal Care and Use Committee and complies with the guidelines of the
American Society of Mammalogists (1998).
58
Cotton mouse models.⎯Using program MARK I ran the best model from the analysis of cotton mouse populations at the Dixon Center (Chapter 1). This model
estimated a unique, monthly survival rate for each interval between primary trapping
periods, but did not distinguish survival between treatments or experimental units. Since
there were no treatments in Conecuh NF and I considered the 3 experimental units to be
samples of the same population, I felt the top model from the Dixon Center analysis was appropriate for Conecuh NF. By using the same model for both locations, I was able to directly compare estimates.
Golden mouse models.⎯To evaluate the effects of time and treatment on golden
mice survival at the Dixon Center I compared an a priori set of candidate models. These
models are similar to those used for the analysis of cotton mouse survival (Chapter 1), but since the data set is much smaller, the underlying model structure is simpler.
A suitable goodness of fit test for the robust design is not currently available. The robust design in program MARK estimates apparent survival (survival and emigration), temporary emigration, initial capture probability, and recapture probability. I set program
MARK to calculate monthly survival rates by defining the 9 intervals between primary
periods in monthly units (e.g., 45 days = 1.5 month). I treated each experimental unit as a
separate group. On 2 of the experimental units, only 1 golden mouse was ever captured.
I did not include these 2 units in the modeling procedure, leaving 13 groups.
Data were too sparse to support complex models with numerous parameters.
Since my interest was in survival, I used the same structure of temporary emigration,
capture, and recapture probabilities for all models. I assumed that animals did not
temporarily emigrate from the study area. This is a reasonable assumption because
59
golden mice are sedentary and have a home range < 0.6 ha (Linzey and Packard 1977). I set the model to calculate one estimate of capture probability for each primary trapping
period. I modeled a trap response within primary periods by setting recapture
probabilities equal to initial capture probabilities plus an additive behavioral response. I
modeled a trap response across primary periods by using the Huggins estimator (Huggins
1989, 1991) to include previous capture history as a covariate with capture probability. I
defined previous capture history with a dummy variable indicating whether the individual
had been caught in the previous primary period.
To determine whether survival varied temporally, I fit models with periodic and seasonal variation and a model with no temporal variability. The periodic model (period) estimated the same survival rate for the first 2 intervals between primary periods (due to sparse data) and estimated a unique survival rate for each period thereafter. The seasonal model (season) estimated a unique survival rate for each of 3 seasons defined by the time of year between primary periods – spring, summer, or fall-winter. Seasonal estimates did not vary between years.
I modeled treatments in three ways: 1) I handled each treatment as a separate
effect on survival (trt). The 2 units that were removed from modeling were both
thin/burn units. I treated the remaining thin/burn unit as a burn unit. 2) I constrained
survival rates on thin units to be equal to those of control units, while the 2 other treatments were distinguished (thin=control), and 3) I constrained all units that were
burned to have equal survival rates, while survival rates on thin units were equal to those
of control units (burn). Because this later model only used 1 term for the treatment
60
effect, it was possible to model an interaction between time and treatment (period*burn)
without creating a model that contained more parameters than the data could support.
For all models, I constrained estimates of survival to be equal across all sites for
the pre-treatment trapping periods. I added each of the 3 parameterizations of treatment
effects to the period, season and null construction of time. To account for variation
between units within treatments, I added a unit effect. I also included temporal models
with no treatment effects, and a model with no variation in survival rates. All
combinations of time, treatment, and unit effects constitute the candidate model set.
Model selection and inference followed the methodology of Burnham and
Anderson (2002). I ranked models using Akaike’s Information Criterion corrected for
small sample size (AICc; Akaike 1973). The model with the lowest AICc value is the
most parsimonious approximation of truth among the set of candidate models, given the
data. I used model deviance, the difference in AICc values between the top ranked model
and each competing model (ΔAICc), and normalized AICc values (Akaike weights, wi) to compare the relative performance of competing models. Akaike weights can be viewed as the weight of evidence in favor of a particular model. The ratio of Akaike weights
(wi/wj), or evidence ratio, indicates the likelihood of one model over another (Burnham
and Anderson 2002).
RESULTS
Community structure and membership.⎯Ten species were caught at the Dixon
Center and 7 species were caught in Conecuh NF (Table 1). Cotton mice were the most
abundant species at both locations. Golden mice and southern flying squirrels
(Glaucomys volans) were the second and third most frequently captured species in
61
woodland while cotton rats and eastern harvest mice were the second and third most
frequently captured species in savanna. Four species caught at the Dixon Center were not
caught in Compartment 58 – southern flying squirrel, eastern woodrat (Neotoma
floridana), eastern chipmunk (Tamius striatus), and pine vole (microtus pinetorum).
Only one species caught in Compartment 58 was not caught at the Dixon Center – eastern harvest mouse (Reithrodontomys humulis). No oldfield mice (P. polionotus) were captured.
Cotton mice.⎯Estimated rates of survival for cotton mice in savanna habitat
fluctuated much more widely than those for cotton mice living in woodland habitat (Fig.
2; Chapter 1). Fall survival rates were substantially lower in savanna habitat than in
woodland habitat. Spring survival in the savanna exceeded that of woodlands. Summer
survival was also higher in savanna than woodlands, but confidence intervals between the
estimates overlapped. Estimates of abundance ranged from 0 (no individuals captured) to
55.05 (95% CI = 16.20, 228.58). Wide 95% confidence intervals made estimates of
abundance indistinguishable from those of populations at the Dixon Center.
Golden mice.⎯Judging by ΔAICc, AICc weights, and evidence ratios, any of the
top three models could have been the one that best approximates truth, given the set of
candidate models (Table 2; Burnham and Anderson 2002). The top two models included
the burn treatment structure – all units that were burned were constrained to have the
same estimate of survival while thin units were constrained to have the same estimates of
survival as control units. These two models hold 62.5% of the AICc weight. The third
ranked model allowed no variation in survival estimates. While the fourth and fifth
models had a ΔAICc < 4, they had low weights and the evidence ratio indicated they did
62
not compare favorably with the top model. The fifth model had a deviance nearly identical to the null model but a larger AICc value, indicating the thin=control
parameterization added no information.
The estimated effect of burning was a 13% reduction in monthly survival rates of
golden mice (model burn β = -0.56, (95% CI = -1.11, 0.0004; model season + burn β =
-0.53, 95% CI = -1.08, 0.01). The top ranked model, burn, estimated survival to be 0.72
(95% CI = 0.66, 0.77) on unburned units and 0.59 (95% CI = 0.47, 0.70) on burned units.
Confidence intervals on estimates from model season + burn, the second ranked model,
overlapped between burned and unburned units (Fig. 3). The estimated survival rate
from the model containing no variation was 0.69 (95% CI = 0.63, 0.74).
The initial probability of catching a golden mouse was low (for model burn 0.04
to 0.16, 95% CI = 0.02, 0.11 and 0.08, 0.30, respectively). Golden mice responded
positively to trapping, both within and between primary periods. Recapture rate was 35%
higher than the probability of initial capture (β = 1.55, 95% CI = 1.36, 1.73). Individuals
that had been captured at time i-1 were 32% more likely to be captured at time i (β =
1.41, 95% CI = 1.18, 1.63).
Estimates of golden mouse abundance from the top model ranged from 0.00 to
43.10 (95% CI = 24.59, 97.31). Golden mice were captured before treatment on 6 of 9 units to be burned. They were eradicated from 2 sites after burning – a thin/burn unit and a herbicide/burn unit. The herbicide/burn unit held the largest estimated population of all golden mice. This occurred in the winter of 2003, after herbicide was applied but before the site was burned. Golden mice were only captured on 2 of 6 thin and control units prior to treatment. Through the course of the study, I caught golden mice on all of the
63
other thin and control units. Conversely, on 3 burned units I never caught golden mice,
before or after treatment. Abundance estimates were too imprecise to distinguish
between treatments or between pre-treatment and post-treatment periods.
Cotton rats.⎯Only one cotton rat was caught in more than one primary period.
This individual was caught at the Dixon Center on a burn unit in March and June of 2004.
This is the unit that was burned late in 2004. No cotton rat was caught in July 2004, after the second burning of the unit.
DISCUSSION
Community structure and membership.⎯The woodland community differs
from the savanna community in both structure and membership. The habitat generalist,
the cotton mouse, dominates both. Other members of these communities are more
specialized in habitat preference. Members of the woodland community require
components found there – flying squirrels require nest cavities in deciduous trees, golden
mice prefer thick tangles of vines and shrubs. The savanna community is filled with
species that specialize in grassland habitat – the cotton rat and the eastern harvest mouse.
Abundance appeared to be more evenly proportioned between species in the savanna, but
this observation is based on the number of individuals captured for each species. Besides
the cotton mouse and the golden mouse, too few individuals were captured to calculate
abundances. The southern flying squirrel was probably under-represented because traps
were placed on the ground. The trapping method was also inadequate to sample shrew
populations.
64
Cotton mice.⎯Support was mixed for my hypothesis that frequently burned savanna is of higher quality for cotton mice than woodlands. Survival rates of cotton mice living in savanna were higher in the spring and summer, but lower in the fall, than rates for mice living in woodlands. The thin/burn and herbicide/burn treatments at the
Dixon Center raised survival rates for cotton mice, indicating the beneficial effects of fire improved habitat (Chapter 1), which is consistent with higher spring and summer survival rates in the savanna. One possible explanation for low survival rates in savanna in the
fall is the absence of hardwoods and a hard mast crop in the fall. At least 3 publications
have noted a correlation between acorn production and an increase in the cotton mouse
population (McCarley 1954, Pearson 1953, Stout 1976). Without a better understanding
of low fall survival rates and mechanisms causing fluctuations in abundance, the relative
quality of the savanna habitat is uncertain.
In the only other publication of survival rates for the cotton mouse of which I am
aware, Sasso and Gaines (2002) used the Jolly-Seber method (Jolly 1965, Seber 1965) to
estimate mean 28-day survival rates for Key Largo cotton mice (P. g. allapaticola) living
in 4 age classes of hammock. They reported no temporal variation in survival rates.
Hydric and mesic hammocks, bottomland hardwood forests, and swamps (bottomland
hardwoods) are purported to be optimal habitat for cotton mice (Wolfe and Linzey 1977).
This assertion has not been supported by any demographic data other than abundance.
The 4 mean estimates reported by Sasso and Gaines (0.68 ± 0.04, 0.71 ± 0.05, 0.74 ±
0.03, 0.78 ± 0.05; 2002) all fall below the spring and summer estimates of survival for
adults in Compartment 58 of Conecuh NF. This contradicts the assumption that
bottomland hardwoods are optimal habitat for cotton mice. In Chapter 1, I cite evidence
65
of emigration from bottomland hardwoods to upland experimental units as circumstantial
support for the hypothesis that bottomland hardwoods are source habitat for cotton mice.
This apparent conflict begs more thorough sampling in both bottomland hardwoods and
longleaf pine savannas. I sampled only 1 forest compartment in Conecuh NF; Sasso and
Gaines (2002) sampled 4 hammocks, all on Key Largo. Sasso and Gaines (2002) did not distinguish between adult and juvenile survival, which may have resulted in slightly
lower estimates. I found juvenile survival rates to be approximately 9% less than adult
survival rates (Chapter 1).
A large portion of the cotton mouse range (Wolf and Linzey 1977) is concurrent
with the former range of longleaf pine (Frost 1993), suggesting a common evolutionary
history. Habitat available to cotton mice in the pre-settlement landscape of North
America would have been hardwoods in low-lying areas and longleaf pine in the upland
(Platt 1999). Both bottomland hardwood forests and longleaf pine savannas are subject
to periodic disturbance (floods and fire, respectively), which may have caused cotton
mice to frequently move between upland and lowland. I hypothesize optimal habitat for
cotton mice is a composite of bottomland hardwood forests and longleaf pine. I suggest
that landscapes, not habitat “patches,” are the scale of importance to cotton mice and the
appropriate perspective for restoration of the longleaf pine ecosystem.
Golden mice.⎯Model selection results support the hypothesis that fire reduces
golden mouse survival. The 13% reduction in monthly survival rates equates to a 90%
annual reduction. This reduction could have been due to mortality or individuals leaving
the trapping area. It appears that golden mice colonized thin and control units but were
prevented from colonizing burned units. This observation may only be a spurious effect
66
caused by the extremely low probability of catching a golden mouse for the first time. It
is difficult to further assess the effects of fire on golden mice abundance because
estimates were imprecise. The high ranking of the null model indicates that the data were
too sparse to support complex models of the structure of survival.
Cotton rats.⎯The hypothesis that burning would create favorable habitat for
cotton rats was not supported. Other than the 1 cotton rat that took up residence on a
burned unit, cotton rats did not immigrate onto units that were burned. Grasses on this
site did expand in coverage after initial treatment, but they were eliminated after the
second burn (personal observation), which probably explains why the cotton rat was not
captured during the final trapping period.
Management implications.⎯The goal of ecosystem restoration is to re-instate a
particular community to a location where it was known or thought to have previously
existed before anthropogenic disturbance. Implicit in this goal is the concept that
ecosystems are unique by virtue of inherent biotic and abiotic processes and each
member of the community plays some role in these processes.
Small mammals can play a significant role in shaping the ecosystem by
consuming seeds and arthropods and by disturbing the soil (Boyer 1964, Grant and
French 1980, Brown and Heske 1990, Ostfeld et al. 1997, Talmon et al. 2003).
Ecosystem function can be affected by changes in density or composition of the small
mammal community (Grant and French 1980). Brown and Heske (1990) showed that 3
species of kangaroo rats (Dipodomys spp.) acting as a keystone guild were essential to
prevent desert shrubland from converting to grassland. Restoring the small mammal community can be pivotal to restoring ecosystem function.
67
The community at the Dixon Center will have to undergo substantial change
before it resembles the community in Conecuh NF. No changes in community
membership due to treatments were observed; golden mice were not eradicated and
cotton rats did not immigrate. Lowered survival of golden mice indicates changes at the population level have begun. Returning regular, spring fires to longleaf pine stands should re-instate the process under which this ecosystem evolved and thereby a community that persists under this process. Since I was not able to distinguish a difference in golden mouse survival between the 3 treatments that included burning, I cannot recommend which treatment is best. Results from Chapter 1 indicate the combination treatments are most effective. It may require several episodes of burning before substantial changes in the community take place. If the process is continued, eventually species preferring woodland habitat may retreat to bottomland hardwoods.
The result may be an upland community that comprises fewer species.
A picture of the reference community helps define restoration goals and evaluate
the efficacy of restoration efforts. Negative trends in golden mouse survival can be
viewed positively because a species that is not a member of the desired upland
community will probably retreat to the bottomland hardwoods, where it can find the shrubs and vines it prefers. In the context of restoring longleaf pine uplands, the removal of some species is expected. If focus were held only on the uplands, the perceived loss of species could be interpreted negatively as a loss of biodiversity. If the correct perspective
for restoration of the longleaf pine ecosystem is a landscape of upland longleaf pine
interspersed with bottomland hardwoods, as I have suggested, then species richness is not
lost. Rather, some species will relocate. Some ecosystems, such as longleaf pine or pine
68
pocosins (Mitchell et al. 1995), may be naturally poor in small mammal species.
Biodiversity is a poorly defined concept and a legal quagmire (Bosselman 2004).
Management decisions should be based upon an understanding of the ecosystem in which management actions will occur.
ACKNOWLEDGEMENTS
I greatly appreciate the assistance of S. Converse with program MARK and model development. G. White provided helpful suggestions and insights. I am grateful to all the field technicians that helped with the small mammal trapping. J. Christian provided administrative support. I thank the USDA Forest Service for permission to trap in
Conecuh National Forest. This is contribution number XX of the National Fire and Fire
Surrogate Research (FFS) Project. This research was funded by the USDA Forest
Service through the National Fire Plan. Although the author received no direct funding for this research from the U.S. Joint Fire Science Program (JFSP), it was greatly facilitated by the JFSP support of existing FFS project sites.
69
LITERATURE CITED
Animal Care and Use Committee. 1998. Guidelines for the capture, handling, and care of
mammals as approved by The American Society of Mammalogists. Journal of
Mammalogy 79:1416-1431.
Ahlgren, C. E. 1966. Small mammals and reforestation following prescribed burning.
Journal of Forestry 64:614-618.
Arata, A. A. 1959. Effects of burning on vegetation and rodent populations in a longleaf
pine-turkey oak association in north central Florida. Quarterly Journal of the
Florida Academy of Science 22:94-104.
Beck, A. M. and R. J. Vogl. 1972. The effects of spring burning on rodent populations in
a brush prairie savanna. Journal of Mammalogy 53:336-345.
Bosselman, F. 2004. A dozen biodiversity puzzles. New York University Environmental
Law Journal 12:364-506.
Boyer, W. D. 1964. Longleaf seed losses to animals on burned seedbeds. United States
Department of Agriculture, Forest Service, Southern Forestry Experimental
Station, Research Note SO-6:1-3.
Brockway, D. G. and K. W. Outcalt. 2000. Restoring longleaf pine wiregrass ecosystems:
hexazinone application enhances effects of prescribed fire. Forest Ecology and
Management 137:121-138.
Brown, J. H. and E. J. Heske. 1990. Control of a desert-grassland transition by a keystone
rodent guild. Science 250:1705-1707.
70
Burnham, K. P. and D. R. Anderson. 2002. Model selection and inference: a practical
information-theoretic approach. 2nd ed. Springer-Verlag, Inc, New York.
Cameron, G. N. and S. R. Spencer. 1981. Sigmodon hispidus. Mammalian Species 158:1-
9.
Dublin, H. T., A. R. E. Sinclair, and J. McGlade. 1990. Elephants and fire as causes of
multiple stable states in the Serengeti-Mara woodlands. Journal of Animal
Ecology 59:1147-1164.
Fischer, R. A. and N. R. Holler. 1990. Habitat use and relative abundance of gray
squirrels in southern Alabama. Journal of Wildlife Management 55:52-59.
Frost, C. C. 1993. Four centuries of changing landscape patterns in the longleaf pine
ecosystem. Pp. 17-43 in The longleaf pine ecosystem: ecology, restoration and
management. (S.M. Hermann, K. Ross, and K. Gainey eds.). Proceedings of the
18th Tall Timbers fire ecology conference. Tall Timbers Research, Inc.,
Tallahassee, Florida.
Jolly, G. M. 1965. Explicit estimates from capture-recapture data with both death and
immigration-stochastic model. Biometricka 52:225-247.
Grant, W. E. and N. R. French. 1980. Evaluation of the role of small mammals in
grassland ecosystems: a modeling approach. Ecological Modeling 8:15-37.
Komarek, E. V. 1939. A progress report on south-eastern mammal studies. Journal of
Mammalogy 20:292-299.
Landers, J. L. and W. D. Boyer. 1999. An old-growth definition for upland longleaf and
south Florida slash pine forests, woodlands, and savannas. United States
71
Department of Agriculture, Forest Service, Southern Research Station, General
Technical Report SRS-29:1-19.
Layne, J. N. 1974. Ecology of small mammals in a flatwoods habitat in north central
Florida, with special emphasis on the cotton rat (Sigmodon hispidus). American
Museum Novitates 2544:1-48.
Linzey, D. W. and R. L. Packard. 1977. Ochrotomys nuttalli. Mammalian Species 75:1-6.
Masters, R. E., R. L. Lochmiller, S.T. McMurry, and G.A. Bukenhofer. 1998. Small
mammal response to pine-grassland restoration for red-cockaded woodpeckers.
Wildlife Society Bulletin 26:148-158.
McCarley, W. H. 1954. Fluctuations and structure of Peromyscus gossypinus population
in eastern Texas. Journal of Mammalogy 35:526-532.
McMurry, S. T., R. L. Lochmiller, J.F. Boggs, D.M. Leslie, Jr., and D.M. Engle. 1996.
Demography and condition of populations of white-footed mice (Peromyscus
leucopus) in late and early successional habitats. Journal of Mammalogy 77:335-
345.
Mitchell, M. S., K. S. Karriker, E. J. Jones, and R. A. Lancia. 1995. Small mammal
communities associated with pine plantation management of pocosins. Journal of
Wildlife Management 59:875-881.
Ostfeld, R. S., R. H. Manson, and C. D. Canham. 1997. Effects of rodents on survival of
tree seeds and seedlings invading old fields. Ecology 78:1531-1542.
Outcalt, K. W. 2002. Study plan: fire and fire surrogate treatments in the Gulf Coastal
Plain Solon Dixon Forestry and Education Center, Alabama. On file, United
72
States Department of Agriculture, Forest Service, Southern Reasearch Station,
Athens, GA.
Pearson, P. G. 1953. A field study of Peromyscus populations in Gulf Hammock, Florida.
Ecology, 34:199-207.
Peet, R. K. and D. J. Allard. 1993. Longleaf pine vegetation of the southern Atlantic and
eastern Gulf coast regions: A preliminary classification. Pp. 45-82 in The
longleaf pine ecosystem: ecology, restoration and management. (S.M. Hermann,
K. Ross, and K. Gainey eds.). Proceedings of the 18th Tall Timbers fire ecology
conference. Tall Timbers Research, Inc., Tallahassee, Florida.
Pollock, K. H. 1982. A capture-recapture sampling design robust to unequal catchability.
Journal of wildlife management 46:752-757.
Ripple, W. J. and R. L. Beschta. 2004. Wolves and the ecology of fear: can predation risk
structure ecosystems? BioScience 54:755-766.
Sasso, C. R. and M. S. Gaines. 2002. Population ecology of three species of small
mammals on Key Largo, Florida. Florida Scientist 65:115-125.
Seber, G. A. F. 1965. A note on the multiple recapture census. Biometrika 52:249-259.
Stout, I. J. 1976. Response of small mammals in a scrub community to supplementary
food. Florida Science 39 (Supplement 1):8.
Stransky, J. J. and R. F. Harlow. 1981. Effects of fire on deer habitat in the Southeast. Pp.
135-142 in Prescribed fire and wildlife in southern forests: proceedings of the
symposium. (G.W. Wood ed.). The Belle W. Baruch Forest Science Institute,
Clemson University, Georgetown, South Carolina.
73
Sullivan, T. P. and D. S. Sullivan. 1982. Responses of small-mammal populations to a
forest herbicide application in a 20-year-old conifer plantation. Journal of Applied
Ecology 19:95-106.
Tallmon, D. A., E. S. Jules, N. J. Radke, and L. S. Mills. 2003. Of mice and men and
trillium: cascading effects of forest fragmentation. Ecological Applications
13:1193-1203.
Van Lear, D. H and R. F. Harlow. 2000. Fire in the eastern United States: Influence on
wildlife habitat. Pp. 2-10 in The role of fire in non-game wildlife management
and community restoration: traditional uses and new direction. Proceedings of a
special workshop. (W.M. Ford, K.R. Russell, and C.E. Moorman eds.). United
States Department of Agriculture, Forest Service, Northeastern Research Station,
General Technical Report NE-288:1-145.
Weatherspoon, C. P., and J. McIver. 2000. A national study of the consequences of fire
and fire surrogate treatments. United States Department of Agriculture, Forest
Service, Pacific South West Research Station, Proposal to the Joint Fire Science
Program:1-115.
White, G. C. and K. P. Burnham. 1999. Program MARK: survival estimation from
populations of marked animals. Bird Study 46 Supplement:120-138.
Williams, B. K., J. D. Nichols, and M. J. Conroy. 2002. Analysis and management of
animal populations. Academic Press, New York, New York, USA.
Wolfe, J. L. and A. V. Linzey. 1977. Peromyscus gossypinus. Mammalian Species 70:1-
5.
74
TABLES AND FIGURES
Table 1.⎯Total individuals captured over the duration of trapping, in hardwood-pine woodland and pine savanna in Covington and Escambia Counties, Alabama, 2001-2004.
Species # Individuals % of Total
Hardwood-pine woodland
Peromyscus gossypinus 1190 0.84
Ochrotomys nuttalli 148 0.11
Glaucomys volans 47 0.03
Sigmodon hispidus 14 0.01
Neotoma floridana 4 0.00
Oryzomys palustris 3 0.00
Tamias Striatus 3 0.00
Microtus pinetorum *
Blarina carolinensis *
Cryptotis parva *
Total 1409
Pine Savanna
Peromyscus gossypinus 139 0.68
Sigmodon hispidus 33 0.16
Reithrodontomys humulis 27 0.13
Ochrotomys nuttalli 5 0.02
Oryzomys palustris 1 0.00
Blarina carolinensis *
Cryptotis parva *
Total 205 * Species known to be present but uncounted.
75
Table 2.⎯Model selection results for golden mouse survival in longleaf pine stands undergoing treatments for habitat restoration in Covington and Escambia Counties,
Alabama, 2001-2004. Models are ranked in ascending order by Akaike’s Information
Criterion, corrected for small sample size (AICc). Effects of time were modeled seasonally or periodically. For seasonal structure (season) the intervals between primary periods were defined as spring, summer, or fall-winter and one estimate of survival was calculated for each of the three seasons. For periodic survival (period) a unique survival rate was for estimated each of the 9 intervals between primary periods. Treatments were modeled in 1 of 3 ways: 1) each treatment – thin, burn, thin/burn, herbicide/burn, and control – was estimated to have unique effect (trt), 2) thin units were estimated to have the same estimates of survival as control units; all other treatments had a unique effect
(thin=control), 3) all units that were burned were constrained to have the same estimate of survival; thin units were constrained to have the same estimates of survival as control units (burn). Individual effects for each experimental unit were included, representing variation between units within treatments. K = number of parameters. Evidence ratio indicates the likelihood of the top ranked model versus the competing model (e.g., the top ranked model is 2.31 times more likely to be the model that best approximates truth than the second ranked model).
76
Model AICc Δ AICc AICc Weights K Deviance Evidence ratio burn 1938.81 0.00 0.44 14 1909.89 - season + burn 1940.48 1.67 0.19 16 1907.28 2.31 null 1940.61 1.80 0.18 13 1913.81 2.46 season 1942.09 3.28 0.08 15 1911.04 5.17 thin=cont 1942.72 3.91 0.06 14 1913.80 7.07 season + thin=cont 1945.04 6.23 0.02 17 1909.69 period + burn 1946.43 7.62 0.01 21 1902.37 trt 1946.51 7.70 0.01 16 1913.31 season + trt 1947.17 8.36 0.01 18 1909.66 period 1948.23 9.42 0.00 20 1906.36 period*burn 1950.64 11.84 0.00 25 1897.73 period + thin=cont 1951.36 12.55 0.00 22 1905.11 period + trt 1953.47 14.66 0.00 23 1905.00 unit 1958.03 19.23 0.00 24 1907.35 season + unit + burn 1958.34 19.53 0.00 27 1900.94 season + unit + thin=cont 1959.80 21.00 0.00 27 1902.40 season + unit 1960.41 21.60 0.00 26 1905.25 season + unit + trt 1961.72 22.91 0.00 28 1902.06 period + unit + burn 1963.16 24.35 0.00 32 1894.35 period + unit 1966.55 27.74 0.00 31 1900.04 period*burn + unit 1966.86 28.05 0.00 36 1888.74 period + unit + thin=cont 1968.04 29.23 0.00 32 1899.23 period + unit + trt 1970.35 31.54 0.00 33 1899.22
77
Fig. 1.⎯ Treatment and trapping schedule for a replicated experiment and capture- recapture study of the effects of fire and fire alternatives on small mammals in longleaf pine stands, Covington and Escambia Counties, Alabama, 2001-2004.
Fig. 2.⎯ Monthly adult survival rates and 95% confidence intervals, estimated for each interval between ten primary periods, for cotton mice living in hardwood-pine woodlands and pine savannas in Covington and Escambia Counties, Alabama, 2001-2004.
Fig. 3.⎯ Monthly survival rates and 95% confidence intervals, estimated for each of three seasons, for golden mice on burned and unburned experimental units in Covington and Escambia Counties, Alabama, 2001-2004.
78
Fig. 1.
Primary period
12 34 5 67 8 910
Burn herbicide 2nd burn Thin Burn Herbicide units 2nd burn unit 11
1 3 01 02 03 04 0 002 00 004 0 200 20 2 20 2 2 y 2002 ul 2003 n 2 ul 2004 un 2 Jul a un 2002Jul ep ay J ay J J J S Feb 2003 Jun Feb 2004 Ju r-M r-M r-M p Feb-AprAp 2002 A Ap
79
Fig. 2.
1.0
0.9
0.8
0.7
0.6
0.5
0.4
Monthly survival rate Monthly 0.3
0.2 Woodland 0.1 Savanna
0.0 2 03 04 2001 200 2002 20 2003 20 nter nter Wi Spring 2003 Wi Spring 2004 Summer Summer l - Summer l - Summer al al F F
Fall 2001 - Spring 2002
80
Fig. 3.
0.9
0.8
0.7
0.6
0.5 Monthly survival rates
0.4 Unburned Burned
0.3 Summer Fall - Winter Spring
81
CONCLUSION
The goal of my research was to investigate the effects of fire and fire surrogates
(FFS) for ecosystem restoration on the small mammals living in longleaf pine stands in
which the natural process of fire has been interrupted. To do so, I conducted a capture-
recapture study to track changes in species’ populations over time, space, and treatment.
To make a full accounting of these changes, I hoped to quantify the four factors that
determine population size – births, immigration, deaths, and emigration. I encompassed these four vital rates by calculating apparent survival (survival and emigration) and
recruitment (births and immigration). I had hoped to distinguish births and immigration,
but due to the lack of a distinct breeding season, the dispersal habits of juveniles, and too
few data on reproductively active females, I was unable to do this. Despite this
shortcoming, I feel I was able to construct a fairly detailed accounting of the effects of the FFS treatments on small mammals.
I conclude that the FFS treatments did affect small mammals, but only
moderately. Treatments that combine fire with thinning or herbicide were most effective
for ecosystem restoration. A treatment that included all three agents would probably be
the most expeditious. Treatments should be applied over a larger area. In the early
stages of restoration this may be difficult due to high fuel levels, but eventually fire should be carried out on a scale that mimics the fires that once burned over kilometers of longleaf pine savannas.
82
The small mammal community living in the hardwood-pine woodland stands at
the Dixon Center differs in membership and structure from the community living in the
longleaf pine savannas of Conecuh NF. Several rounds of restoration treatments may be
required before changes in the community are recognized. The species found in the
community are determined, in part, by the pool of species available for immigration from
areas surrounding the restoration area.
The concept of defining habitat for a small mammal over multiple scales, some of
which are quite broad, is not one that is often considered. Innumerable studies have
investigated small mammal interactions with microhabitat. While these natural history
studies are certainly important, my findings suggest that conservation of populations may
require scientists to enlarge the window through which they view small mammals.
My research raised several questions that should be addressed with future
research. While I learned much about cotton mouse population demographics,
information is still lacking about reproductive success in different habitats. Pournelle
(1952) conducted the only detailed study of cotton mouse ontogeny and reproduction.
Information from various habitats on litter size, neonate survival, and age at maturity
would greatly enhance our understanding of cotton mouse populations.
More demographic studies are needed of cotton mice in longleaf pine savannas
and bottomland hardwood forests. I believe that the cotton mouse’s evolution as a habitat
generalist may be linked to its history living in two intertwined, frequently disturbed, and
therefore constantly changed environments.
Finally, my research has raised questions about the correct way to perform a reverse time analysis using the robust design. Typically, reverse time analysis is
83
performed simply by reversing the capture history. This may not be appropriate for the
robust design, which contains secondary sampling periods within the primary periods. I
believe the secondary periods are asymmetrical regarding time, and the appropriate approach to reverse time modeling in the robust design is to reverse the primary periods, but retain the forward time order of the secondary periods. Reverse time modeling is a
relatively new tool in capture-recapture modeling and further consideration of the theoretical underpinnings regarding robust design modeling is needed.
84